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(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 *, u32 pid, 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;
283 int stats_counter_off;
286 OVS_CB(skb)->vport = p;
288 if (!OVS_CB(skb)->flow) {
289 struct sw_flow_key key;
293 /* Extract flow from 'skb' into 'key'. */
294 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
295 if (unlikely(error)) {
300 if (is_frag && dp->drop_frags) {
302 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
307 flow = flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
308 if (unlikely(!flow)) {
309 struct dp_upcall_info upcall;
311 upcall.cmd = OVS_PACKET_CMD_MISS;
313 dp_upcall(dp, skb, &upcall);
315 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
319 OVS_CB(skb)->flow = flow;
322 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
323 flow_used(OVS_CB(skb)->flow, skb);
324 execute_actions(dp, skb);
327 /* Update datapath statistics. */
329 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
331 write_seqcount_begin(&stats->seqlock);
332 (*(u64 *)((u8 *)stats + stats_counter_off))++;
333 write_seqcount_end(&stats->seqlock);
338 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
340 u16 csum_start, csum_offset;
343 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
344 csum_start -= skb_headroom(skb);
346 skb_copy_bits(skb, 0, to, csum_start);
348 csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
349 skb->len - csum_start, 0);
350 *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
353 static struct genl_family dp_packet_genl_family = {
354 .id = GENL_ID_GENERATE,
355 .hdrsize = sizeof(struct ovs_header),
356 .name = OVS_PACKET_FAMILY,
358 .maxattr = OVS_PACKET_ATTR_MAX
361 int dp_upcall(struct datapath *dp, struct sk_buff *skb,
362 const struct dp_upcall_info *upcall_info)
364 struct sk_buff *segs = NULL;
365 struct dp_stats_percpu *stats;
369 if (OVS_CB(skb)->flow)
370 pid = OVS_CB(skb)->flow->upcall_pid;
372 pid = OVS_CB(skb)->vport->upcall_pid;
379 forward_ip_summed(skb, true);
381 /* Break apart GSO packets into their component pieces. Otherwise
382 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
383 if (skb_is_gso(skb)) {
384 segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
393 err = queue_userspace_packets(dp, pid, skb, upcall_info);
395 struct sk_buff *next;
396 /* Free GSO-segments */
400 } while ((segs = next) != NULL);
410 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
412 write_seqcount_begin(&stats->seqlock);
414 write_seqcount_end(&stats->seqlock);
421 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
422 * 'upcall_info'. There will be only one packet unless we broke up a GSO
425 static int queue_userspace_packets(struct datapath *dp, u32 pid,
427 const struct dp_upcall_info *upcall_info)
431 dp_ifindex = get_dpifindex(dp);
436 struct ovs_header *upcall;
437 struct sk_buff *user_skb; /* to be queued to userspace */
442 err = vlan_deaccel_tag(skb);
446 if (nla_attr_size(skb->len) > USHRT_MAX)
449 len = sizeof(struct ovs_header);
450 len += nla_total_size(skb->len);
451 len += nla_total_size(FLOW_BUFSIZE);
452 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
453 len += nla_total_size(8);
455 user_skb = genlmsg_new(len, GFP_ATOMIC);
459 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
460 0, upcall_info->cmd);
461 upcall->dp_ifindex = dp_ifindex;
463 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
464 flow_to_nlattrs(upcall_info->key, user_skb);
465 nla_nest_end(user_skb, nla);
467 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
468 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
469 upcall_info->userdata);
471 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
472 if (skb->ip_summed == CHECKSUM_PARTIAL)
473 copy_and_csum_skb(skb, nla_data(nla));
475 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
477 err = genlmsg_unicast(&init_net, user_skb, pid);
481 } while ((skb = skb->next));
486 /* Called with genl_mutex. */
487 static int flush_flows(int dp_ifindex)
489 struct flow_table *old_table;
490 struct flow_table *new_table;
493 dp = get_dp(dp_ifindex);
497 old_table = get_table_protected(dp);
498 new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
502 rcu_assign_pointer(dp->table, new_table);
504 flow_tbl_deferred_destroy(old_table);
508 static int validate_actions(const struct nlattr *attr, int depth);
510 static int validate_sample(const struct nlattr *attr, int depth)
512 static const struct nla_policy sample_policy[OVS_SAMPLE_ATTR_MAX + 1] =
514 [OVS_SAMPLE_ATTR_PROBABILITY] = {.type = NLA_U32 },
515 [OVS_SAMPLE_ATTR_ACTIONS] = {.type = NLA_UNSPEC },
517 struct nlattr *a[OVS_SAMPLE_ATTR_MAX + 1];
520 error = nla_parse_nested(a, OVS_SAMPLE_ATTR_MAX, attr, sample_policy);
524 if (!a[OVS_SAMPLE_ATTR_PROBABILITY])
526 if (!a[OVS_SAMPLE_ATTR_ACTIONS])
529 return validate_actions(a[OVS_SAMPLE_ATTR_ACTIONS], (depth + 1));
532 static int validate_actions(const struct nlattr *attr, int depth)
534 const struct nlattr *a;
537 if (depth >= SAMPLE_ACTION_DEPTH)
540 nla_for_each_nested(a, attr, rem) {
541 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
542 [OVS_ACTION_ATTR_OUTPUT] = 4,
543 [OVS_ACTION_ATTR_USERSPACE] = 8,
544 [OVS_ACTION_ATTR_PUSH_VLAN] = 2,
545 [OVS_ACTION_ATTR_POP_VLAN] = 0,
546 [OVS_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
547 [OVS_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
548 [OVS_ACTION_ATTR_SET_NW_SRC] = 4,
549 [OVS_ACTION_ATTR_SET_NW_DST] = 4,
550 [OVS_ACTION_ATTR_SET_NW_TOS] = 1,
551 [OVS_ACTION_ATTR_SET_TP_SRC] = 2,
552 [OVS_ACTION_ATTR_SET_TP_DST] = 2,
553 [OVS_ACTION_ATTR_SET_TUNNEL] = 8,
554 [OVS_ACTION_ATTR_SET_PRIORITY] = 4,
555 [OVS_ACTION_ATTR_POP_PRIORITY] = 0,
557 int type = nla_type(a);
559 /* Match expected attr len for given attr len except for
560 * OVS_ACTION_ATTR_SAMPLE, as it has nested actions list which
561 * is variable size. */
562 if (type > OVS_ACTION_ATTR_MAX ||
563 (nla_len(a) != action_lens[type] &&
564 type != OVS_ACTION_ATTR_SAMPLE))
568 case OVS_ACTION_ATTR_UNSPEC:
571 case OVS_ACTION_ATTR_USERSPACE:
572 case OVS_ACTION_ATTR_POP_VLAN:
573 case OVS_ACTION_ATTR_SET_DL_SRC:
574 case OVS_ACTION_ATTR_SET_DL_DST:
575 case OVS_ACTION_ATTR_SET_NW_SRC:
576 case OVS_ACTION_ATTR_SET_NW_DST:
577 case OVS_ACTION_ATTR_SET_TP_SRC:
578 case OVS_ACTION_ATTR_SET_TP_DST:
579 case OVS_ACTION_ATTR_SET_TUNNEL:
580 case OVS_ACTION_ATTR_SET_PRIORITY:
581 case OVS_ACTION_ATTR_POP_PRIORITY:
582 /* No validation needed. */
585 case OVS_ACTION_ATTR_OUTPUT:
586 if (nla_get_u32(a) >= DP_MAX_PORTS)
590 case OVS_ACTION_ATTR_PUSH_VLAN:
591 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
595 case OVS_ACTION_ATTR_SET_NW_TOS:
596 if (nla_get_u8(a) & INET_ECN_MASK)
600 case OVS_ACTION_ATTR_SAMPLE:
601 err = validate_sample(a, depth);
616 static void clear_stats(struct sw_flow *flow)
620 flow->packet_count = 0;
621 flow->byte_count = 0;
624 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
626 struct ovs_header *ovs_header = info->userhdr;
627 struct nlattr **a = info->attrs;
628 struct sw_flow_actions *acts;
629 struct sk_buff *packet;
630 struct sw_flow *flow;
639 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
640 !a[OVS_PACKET_ATTR_ACTIONS] ||
641 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
644 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], 0);
648 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
649 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
653 skb_reserve(packet, NET_IP_ALIGN);
655 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
657 skb_reset_mac_header(packet);
658 eth = eth_hdr(packet);
660 /* Normally, setting the skb 'protocol' field would be handled by a
661 * call to eth_type_trans(), but it assumes there's a sending
662 * device, which we may not have. */
663 if (ntohs(eth->h_proto) >= 1536)
664 packet->protocol = eth->h_proto;
666 packet->protocol = htons(ETH_P_802_2);
668 /* Build an sw_flow for sending this packet. */
674 err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
678 err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
679 &flow->key.eth.tun_id,
680 a[OVS_PACKET_ATTR_KEY]);
684 flow->hash = flow_hash(&flow->key, key_len);
686 if (a[OVS_PACKET_ATTR_UPCALL_PID])
687 flow->upcall_pid = nla_get_u32(a[OVS_PACKET_ATTR_UPCALL_PID]);
689 flow->upcall_pid = NETLINK_CB(skb).pid;
691 acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
695 rcu_assign_pointer(flow->sf_acts, acts);
697 OVS_CB(packet)->flow = flow;
700 dp = get_dp(ovs_header->dp_ifindex);
705 if (flow->key.eth.in_port < DP_MAX_PORTS)
706 OVS_CB(packet)->vport = get_vport_protected(dp,
707 flow->key.eth.in_port);
709 err = execute_actions(dp, packet);
725 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
726 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
727 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
728 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
729 [OVS_PACKET_ATTR_UPCALL_PID] = { .type = NLA_U32 },
732 static struct genl_ops dp_packet_genl_ops[] = {
733 { .cmd = OVS_PACKET_CMD_EXECUTE,
734 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
735 .policy = packet_policy,
736 .doit = ovs_packet_cmd_execute
740 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
743 struct flow_table *table = get_table_protected(dp);
745 stats->n_flows = flow_tbl_count(table);
747 stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
748 for_each_possible_cpu(i) {
749 const struct dp_stats_percpu *percpu_stats;
750 struct dp_stats_percpu local_stats;
753 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
756 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
757 local_stats = *percpu_stats;
758 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
760 stats->n_frags += local_stats.n_frags;
761 stats->n_hit += local_stats.n_hit;
762 stats->n_missed += local_stats.n_missed;
763 stats->n_lost += local_stats.n_lost;
767 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
768 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
769 [OVS_FLOW_ATTR_UPCALL_PID] = { .type = NLA_U32 },
770 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
771 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
774 static struct genl_family dp_flow_genl_family = {
775 .id = GENL_ID_GENERATE,
776 .hdrsize = sizeof(struct ovs_header),
777 .name = OVS_FLOW_FAMILY,
779 .maxattr = OVS_FLOW_ATTR_MAX
782 static struct genl_multicast_group dp_flow_multicast_group = {
783 .name = OVS_FLOW_MCGROUP
786 /* Called with genl_lock. */
787 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
788 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
790 const int skb_orig_len = skb->len;
791 const struct sw_flow_actions *sf_acts;
792 struct ovs_flow_stats stats;
793 struct ovs_header *ovs_header;
799 sf_acts = rcu_dereference_protected(flow->sf_acts,
800 lockdep_genl_is_held());
802 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
806 ovs_header->dp_ifindex = get_dpifindex(dp);
808 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
810 goto nla_put_failure;
811 err = flow_to_nlattrs(&flow->key, skb);
814 nla_nest_end(skb, nla);
816 NLA_PUT_U32(skb, OVS_FLOW_ATTR_UPCALL_PID, flow->upcall_pid);
818 spin_lock_bh(&flow->lock);
820 stats.n_packets = flow->packet_count;
821 stats.n_bytes = flow->byte_count;
822 tcp_flags = flow->tcp_flags;
823 spin_unlock_bh(&flow->lock);
826 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
829 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
832 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
834 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
835 * this is the first flow to be dumped into 'skb'. This is unusual for
836 * Netlink but individual action lists can be longer than
837 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
838 * The userspace caller can always fetch the actions separately if it
839 * really wants them. (Most userspace callers in fact don't care.)
841 * This can only fail for dump operations because the skb is always
842 * properly sized for single flows.
844 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
846 if (err < 0 && skb_orig_len)
849 return genlmsg_end(skb, ovs_header);
854 genlmsg_cancel(skb, ovs_header);
858 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
860 const struct sw_flow_actions *sf_acts;
863 sf_acts = rcu_dereference_protected(flow->sf_acts,
864 lockdep_genl_is_held());
866 len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
867 len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
868 len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
869 len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
870 len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
871 return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
874 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
875 u32 pid, u32 seq, u8 cmd)
880 skb = ovs_flow_cmd_alloc_info(flow);
882 return ERR_PTR(-ENOMEM);
884 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
889 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
891 struct nlattr **a = info->attrs;
892 struct ovs_header *ovs_header = info->userhdr;
893 struct sw_flow_key key;
894 struct sw_flow *flow;
895 struct sk_buff *reply;
897 struct flow_table *table;
903 if (!a[OVS_FLOW_ATTR_KEY])
905 error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
909 /* Validate actions. */
910 if (a[OVS_FLOW_ATTR_ACTIONS]) {
911 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], 0);
914 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
919 dp = get_dp(ovs_header->dp_ifindex);
924 table = get_table_protected(dp);
925 flow = flow_tbl_lookup(table, &key, key_len);
927 struct sw_flow_actions *acts;
929 /* Bail out if we're not allowed to create a new flow. */
931 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
934 /* Expand table, if necessary, to make room. */
935 if (flow_tbl_need_to_expand(table)) {
936 struct flow_table *new_table;
938 new_table = flow_tbl_expand(table);
939 if (!IS_ERR(new_table)) {
940 rcu_assign_pointer(dp->table, new_table);
941 flow_tbl_deferred_destroy(table);
942 table = get_table_protected(dp);
949 error = PTR_ERR(flow);
955 if (a[OVS_FLOW_ATTR_UPCALL_PID])
956 flow->upcall_pid = nla_get_u32(a[OVS_FLOW_ATTR_UPCALL_PID]);
958 flow->upcall_pid = NETLINK_CB(skb).pid;
960 /* Obtain actions. */
961 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
962 error = PTR_ERR(acts);
964 goto error_free_flow;
965 rcu_assign_pointer(flow->sf_acts, acts);
967 /* Put flow in bucket. */
968 flow->hash = flow_hash(&key, key_len);
969 flow_tbl_insert(table, flow);
971 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
972 info->snd_seq, OVS_FLOW_CMD_NEW);
974 /* We found a matching flow. */
975 struct sw_flow_actions *old_acts;
977 /* Bail out if we're not allowed to modify an existing flow.
978 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
979 * because Generic Netlink treats the latter as a dump
980 * request. We also accept NLM_F_EXCL in case that bug ever
984 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
985 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
988 /* Update actions. */
989 old_acts = rcu_dereference_protected(flow->sf_acts,
990 lockdep_genl_is_held());
991 if (a[OVS_FLOW_ATTR_ACTIONS] &&
992 (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
993 memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
994 old_acts->actions_len))) {
995 struct sw_flow_actions *new_acts;
997 new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
998 error = PTR_ERR(new_acts);
999 if (IS_ERR(new_acts))
1002 rcu_assign_pointer(flow->sf_acts, new_acts);
1003 flow_deferred_free_acts(old_acts);
1006 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1007 info->snd_seq, OVS_FLOW_CMD_NEW);
1009 if (a[OVS_FLOW_ATTR_UPCALL_PID])
1010 flow->upcall_pid = nla_get_u32(a[OVS_FLOW_ATTR_UPCALL_PID]);
1013 if (a[OVS_FLOW_ATTR_CLEAR]) {
1014 spin_lock_bh(&flow->lock);
1016 spin_unlock_bh(&flow->lock);
1021 genl_notify(reply, genl_info_net(info), info->snd_pid,
1022 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1024 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1025 dp_flow_multicast_group.id, PTR_ERR(reply));
1034 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1036 struct nlattr **a = info->attrs;
1037 struct ovs_header *ovs_header = info->userhdr;
1038 struct sw_flow_key key;
1039 struct sk_buff *reply;
1040 struct sw_flow *flow;
1041 struct datapath *dp;
1042 struct flow_table *table;
1046 if (!a[OVS_FLOW_ATTR_KEY])
1048 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1052 dp = get_dp(ovs_header->dp_ifindex);
1056 table = get_table_protected(dp);
1057 flow = flow_tbl_lookup(table, &key, key_len);
1061 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
1063 return PTR_ERR(reply);
1065 return genlmsg_reply(reply, info);
1068 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1070 struct nlattr **a = info->attrs;
1071 struct ovs_header *ovs_header = info->userhdr;
1072 struct sw_flow_key key;
1073 struct sk_buff *reply;
1074 struct sw_flow *flow;
1075 struct datapath *dp;
1076 struct flow_table *table;
1080 if (!a[OVS_FLOW_ATTR_KEY])
1081 return flush_flows(ovs_header->dp_ifindex);
1082 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1086 dp = get_dp(ovs_header->dp_ifindex);
1090 table = get_table_protected(dp);
1091 flow = flow_tbl_lookup(table, &key, key_len);
1095 reply = ovs_flow_cmd_alloc_info(flow);
1099 flow_tbl_remove(table, flow);
1101 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1102 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1105 flow_deferred_free(flow);
1107 genl_notify(reply, genl_info_net(info), info->snd_pid,
1108 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1112 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1114 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1115 struct datapath *dp;
1117 dp = get_dp(ovs_header->dp_ifindex);
1122 struct sw_flow *flow;
1125 bucket = cb->args[0];
1127 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1131 if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1132 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1133 OVS_FLOW_CMD_NEW) < 0)
1136 cb->args[0] = bucket;
1142 static struct genl_ops dp_flow_genl_ops[] = {
1143 { .cmd = OVS_FLOW_CMD_NEW,
1144 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1145 .policy = flow_policy,
1146 .doit = ovs_flow_cmd_new_or_set
1148 { .cmd = OVS_FLOW_CMD_DEL,
1149 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1150 .policy = flow_policy,
1151 .doit = ovs_flow_cmd_del
1153 { .cmd = OVS_FLOW_CMD_GET,
1154 .flags = 0, /* OK for unprivileged users. */
1155 .policy = flow_policy,
1156 .doit = ovs_flow_cmd_get,
1157 .dumpit = ovs_flow_cmd_dump
1159 { .cmd = OVS_FLOW_CMD_SET,
1160 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1161 .policy = flow_policy,
1162 .doit = ovs_flow_cmd_new_or_set,
1166 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1167 #ifdef HAVE_NLA_NUL_STRING
1168 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1170 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1171 [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1174 static struct genl_family dp_datapath_genl_family = {
1175 .id = GENL_ID_GENERATE,
1176 .hdrsize = sizeof(struct ovs_header),
1177 .name = OVS_DATAPATH_FAMILY,
1179 .maxattr = OVS_DP_ATTR_MAX
1182 static struct genl_multicast_group dp_datapath_multicast_group = {
1183 .name = OVS_DATAPATH_MCGROUP
1186 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1187 u32 pid, u32 seq, u32 flags, u8 cmd)
1189 struct ovs_header *ovs_header;
1193 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1198 ovs_header->dp_ifindex = get_dpifindex(dp);
1201 err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1204 goto nla_put_failure;
1206 nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1208 goto nla_put_failure;
1209 get_dp_stats(dp, nla_data(nla));
1211 NLA_PUT_U32(skb, OVS_DP_ATTR_IPV4_FRAGS,
1212 dp->drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO);
1214 return genlmsg_end(skb, ovs_header);
1217 genlmsg_cancel(skb, ovs_header);
1222 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1225 struct sk_buff *skb;
1228 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1230 return ERR_PTR(-ENOMEM);
1232 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1235 return ERR_PTR(retval);
1240 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1242 if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
1243 u32 frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
1245 if (frags != OVS_DP_FRAG_ZERO && frags != OVS_DP_FRAG_DROP)
1249 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1252 /* Called with genl_mutex and optionally with RTNL lock also. */
1253 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1255 struct datapath *dp;
1257 if (!a[OVS_DP_ATTR_NAME])
1258 dp = get_dp(ovs_header->dp_ifindex);
1260 struct vport *vport;
1263 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1264 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1267 return dp ? dp : ERR_PTR(-ENODEV);
1270 /* Called with genl_mutex. */
1271 static void change_datapath(struct datapath *dp, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1273 if (a[OVS_DP_ATTR_IPV4_FRAGS])
1274 dp->drop_frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]) == OVS_DP_FRAG_DROP;
1277 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1279 struct nlattr **a = info->attrs;
1280 struct vport_parms parms;
1281 struct sk_buff *reply;
1282 struct datapath *dp;
1283 struct vport *vport;
1287 if (!a[OVS_DP_ATTR_NAME])
1290 err = ovs_dp_cmd_validate(a);
1296 if (!try_module_get(THIS_MODULE))
1297 goto err_unlock_rtnl;
1300 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1302 goto err_put_module;
1303 INIT_LIST_HEAD(&dp->port_list);
1305 /* Initialize kobject for bridge. This will be added as
1306 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1307 dp->ifobj.kset = NULL;
1308 kobject_init(&dp->ifobj, &dp_ktype);
1310 /* Allocate table. */
1312 rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1317 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1318 if (!dp->stats_percpu) {
1320 goto err_destroy_table;
1323 change_datapath(dp, a);
1325 /* Set up our datapath device. */
1326 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1327 parms.type = OVS_VPORT_TYPE_INTERNAL;
1328 parms.options = NULL;
1330 parms.port_no = OVSP_LOCAL;
1331 if (a[OVS_DP_ATTR_UPCALL_PID])
1332 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1334 parms.upcall_pid = NETLINK_CB(skb).pid;
1336 vport = new_vport(&parms);
1337 if (IS_ERR(vport)) {
1338 err = PTR_ERR(vport);
1342 goto err_destroy_percpu;
1345 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1346 err = PTR_ERR(reply);
1348 goto err_destroy_local_port;
1350 list_add_tail(&dp->list_node, &dps);
1351 dp_sysfs_add_dp(dp);
1355 genl_notify(reply, genl_info_net(info), info->snd_pid,
1356 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1359 err_destroy_local_port:
1360 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1362 free_percpu(dp->stats_percpu);
1364 flow_tbl_destroy(get_table_protected(dp));
1368 module_put(THIS_MODULE);
1375 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1377 struct vport *vport, *next_vport;
1378 struct sk_buff *reply;
1379 struct datapath *dp;
1382 err = ovs_dp_cmd_validate(info->attrs);
1387 dp = lookup_datapath(info->userhdr, info->attrs);
1392 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
1393 err = PTR_ERR(reply);
1397 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1398 if (vport->port_no != OVSP_LOCAL)
1399 dp_detach_port(vport);
1401 dp_sysfs_del_dp(dp);
1402 list_del(&dp->list_node);
1403 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1405 /* rtnl_unlock() will wait until all the references to devices that
1406 * are pending unregistration have been dropped. We do it here to
1407 * ensure that any internal devices (which contain DP pointers) are
1408 * fully destroyed before freeing the datapath.
1412 call_rcu(&dp->rcu, destroy_dp_rcu);
1413 module_put(THIS_MODULE);
1415 genl_notify(reply, genl_info_net(info), info->snd_pid,
1416 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1426 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1428 struct sk_buff *reply;
1429 struct datapath *dp;
1432 err = ovs_dp_cmd_validate(info->attrs);
1436 dp = lookup_datapath(info->userhdr, info->attrs);
1440 change_datapath(dp, info->attrs);
1442 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1443 if (IS_ERR(reply)) {
1444 err = PTR_ERR(reply);
1445 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1446 dp_datapath_multicast_group.id, err);
1450 genl_notify(reply, genl_info_net(info), info->snd_pid,
1451 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1455 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1457 struct sk_buff *reply;
1458 struct datapath *dp;
1461 err = ovs_dp_cmd_validate(info->attrs);
1465 dp = lookup_datapath(info->userhdr, info->attrs);
1469 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1471 return PTR_ERR(reply);
1473 return genlmsg_reply(reply, info);
1476 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1478 struct datapath *dp;
1479 int skip = cb->args[0];
1482 list_for_each_entry (dp, &dps, list_node) {
1485 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1486 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1487 OVS_DP_CMD_NEW) < 0)
1497 static struct genl_ops dp_datapath_genl_ops[] = {
1498 { .cmd = OVS_DP_CMD_NEW,
1499 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1500 .policy = datapath_policy,
1501 .doit = ovs_dp_cmd_new
1503 { .cmd = OVS_DP_CMD_DEL,
1504 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1505 .policy = datapath_policy,
1506 .doit = ovs_dp_cmd_del
1508 { .cmd = OVS_DP_CMD_GET,
1509 .flags = 0, /* OK for unprivileged users. */
1510 .policy = datapath_policy,
1511 .doit = ovs_dp_cmd_get,
1512 .dumpit = ovs_dp_cmd_dump
1514 { .cmd = OVS_DP_CMD_SET,
1515 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1516 .policy = datapath_policy,
1517 .doit = ovs_dp_cmd_set,
1521 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1522 #ifdef HAVE_NLA_NUL_STRING
1523 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1524 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1525 [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1527 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1528 [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1530 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1531 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1532 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1533 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1536 static struct genl_family dp_vport_genl_family = {
1537 .id = GENL_ID_GENERATE,
1538 .hdrsize = sizeof(struct ovs_header),
1539 .name = OVS_VPORT_FAMILY,
1541 .maxattr = OVS_VPORT_ATTR_MAX
1544 struct genl_multicast_group dp_vport_multicast_group = {
1545 .name = OVS_VPORT_MCGROUP
1548 /* Called with RTNL lock or RCU read lock. */
1549 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1550 u32 pid, u32 seq, u32 flags, u8 cmd)
1552 struct ovs_header *ovs_header;
1557 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1562 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1564 NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1565 NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1566 NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1567 NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1569 nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats));
1571 goto nla_put_failure;
1573 vport_get_stats(vport, nla_data(nla));
1575 NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1577 err = vport_get_options(vport, skb);
1578 if (err == -EMSGSIZE)
1581 ifindex = vport_get_ifindex(vport);
1583 NLA_PUT_U32(skb, OVS_VPORT_ATTR_IFINDEX, ifindex);
1585 return genlmsg_end(skb, ovs_header);
1590 genlmsg_cancel(skb, ovs_header);
1594 /* Called with RTNL lock or RCU read lock. */
1595 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1598 struct sk_buff *skb;
1601 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1603 return ERR_PTR(-ENOMEM);
1605 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1608 return ERR_PTR(retval);
1613 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1615 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1618 /* Called with RTNL lock or RCU read lock. */
1619 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1620 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1622 struct datapath *dp;
1623 struct vport *vport;
1625 if (a[OVS_VPORT_ATTR_NAME]) {
1626 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1628 return ERR_PTR(-ENODEV);
1630 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1631 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1633 if (port_no >= DP_MAX_PORTS)
1634 return ERR_PTR(-EFBIG);
1636 dp = get_dp(ovs_header->dp_ifindex);
1638 return ERR_PTR(-ENODEV);
1640 vport = get_vport_protected(dp, port_no);
1642 return ERR_PTR(-ENOENT);
1645 return ERR_PTR(-EINVAL);
1648 /* Called with RTNL lock. */
1649 static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1653 if (a[OVS_VPORT_ATTR_STATS])
1654 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1656 if (a[OVS_VPORT_ATTR_ADDRESS])
1657 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1662 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1664 struct nlattr **a = info->attrs;
1665 struct ovs_header *ovs_header = info->userhdr;
1666 struct vport_parms parms;
1667 struct sk_buff *reply;
1668 struct vport *vport;
1669 struct datapath *dp;
1674 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE])
1677 err = ovs_vport_cmd_validate(a);
1682 dp = get_dp(ovs_header->dp_ifindex);
1687 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1688 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1691 if (port_no >= DP_MAX_PORTS)
1694 vport = get_vport_protected(dp, port_no);
1699 for (port_no = 1; ; port_no++) {
1700 if (port_no >= DP_MAX_PORTS) {
1704 vport = get_vport_protected(dp, port_no);
1710 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1711 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1712 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1714 parms.port_no = port_no;
1715 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1716 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1718 parms.upcall_pid = NETLINK_CB(skb).pid;
1720 vport = new_vport(&parms);
1721 err = PTR_ERR(vport);
1725 dp_sysfs_add_if(vport);
1727 err = change_vport(vport, a);
1729 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1730 info->snd_seq, OVS_VPORT_CMD_NEW);
1732 err = PTR_ERR(reply);
1735 dp_detach_port(vport);
1738 genl_notify(reply, genl_info_net(info), info->snd_pid,
1739 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1748 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1750 struct nlattr **a = info->attrs;
1751 struct sk_buff *reply;
1752 struct vport *vport;
1755 err = ovs_vport_cmd_validate(a);
1760 vport = lookup_vport(info->userhdr, a);
1761 err = PTR_ERR(vport);
1766 if (a[OVS_VPORT_ATTR_OPTIONS])
1767 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1769 err = change_vport(vport, a);
1770 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1771 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1773 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1775 if (IS_ERR(reply)) {
1776 err = PTR_ERR(reply);
1777 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1778 dp_vport_multicast_group.id, err);
1782 genl_notify(reply, genl_info_net(info), info->snd_pid,
1783 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1791 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1793 struct nlattr **a = info->attrs;
1794 struct sk_buff *reply;
1795 struct vport *vport;
1798 err = ovs_vport_cmd_validate(a);
1803 vport = lookup_vport(info->userhdr, a);
1804 err = PTR_ERR(vport);
1808 if (vport->port_no == OVSP_LOCAL) {
1813 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1815 err = PTR_ERR(reply);
1819 dp_detach_port(vport);
1821 genl_notify(reply, genl_info_net(info), info->snd_pid,
1822 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1830 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1832 struct nlattr **a = info->attrs;
1833 struct ovs_header *ovs_header = info->userhdr;
1834 struct sk_buff *reply;
1835 struct vport *vport;
1838 err = ovs_vport_cmd_validate(a);
1843 vport = lookup_vport(ovs_header, a);
1844 err = PTR_ERR(vport);
1848 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1850 err = PTR_ERR(reply);
1856 return genlmsg_reply(reply, info);
1864 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1866 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1867 struct datapath *dp;
1871 dp = get_dp(ovs_header->dp_ifindex);
1876 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1877 struct vport *vport;
1879 vport = get_vport_protected(dp, port_no);
1883 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1884 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1885 OVS_VPORT_CMD_NEW) < 0)
1890 cb->args[0] = port_no;
1896 static struct genl_ops dp_vport_genl_ops[] = {
1897 { .cmd = OVS_VPORT_CMD_NEW,
1898 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1899 .policy = vport_policy,
1900 .doit = ovs_vport_cmd_new
1902 { .cmd = OVS_VPORT_CMD_DEL,
1903 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1904 .policy = vport_policy,
1905 .doit = ovs_vport_cmd_del
1907 { .cmd = OVS_VPORT_CMD_GET,
1908 .flags = 0, /* OK for unprivileged users. */
1909 .policy = vport_policy,
1910 .doit = ovs_vport_cmd_get,
1911 .dumpit = ovs_vport_cmd_dump
1913 { .cmd = OVS_VPORT_CMD_SET,
1914 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1915 .policy = vport_policy,
1916 .doit = ovs_vport_cmd_set,
1920 struct genl_family_and_ops {
1921 struct genl_family *family;
1922 struct genl_ops *ops;
1924 struct genl_multicast_group *group;
1927 static const struct genl_family_and_ops dp_genl_families[] = {
1928 { &dp_datapath_genl_family,
1929 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1930 &dp_datapath_multicast_group },
1931 { &dp_vport_genl_family,
1932 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1933 &dp_vport_multicast_group },
1934 { &dp_flow_genl_family,
1935 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1936 &dp_flow_multicast_group },
1937 { &dp_packet_genl_family,
1938 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1942 static void dp_unregister_genl(int n_families)
1946 for (i = 0; i < n_families; i++)
1947 genl_unregister_family(dp_genl_families[i].family);
1950 static int dp_register_genl(void)
1957 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1958 const struct genl_family_and_ops *f = &dp_genl_families[i];
1960 err = genl_register_family_with_ops(f->family, f->ops,
1967 err = genl_register_mc_group(f->family, f->group);
1976 dp_unregister_genl(n_registered);
1980 static int __init dp_init(void)
1982 struct sk_buff *dummy_skb;
1985 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1987 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
1995 goto error_tnl_exit;
1999 goto error_flow_exit;
2001 err = register_netdevice_notifier(&dp_device_notifier);
2003 goto error_vport_exit;
2005 err = dp_register_genl();
2007 goto error_unreg_notifier;
2011 error_unreg_notifier:
2012 unregister_netdevice_notifier(&dp_device_notifier);
2023 static void dp_cleanup(void)
2026 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2027 unregister_netdevice_notifier(&dp_device_notifier);
2033 module_init(dp_init);
2034 module_exit(dp_cleanup);
2036 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2037 MODULE_LICENSE("GPL");