blob: 484b7c2fa21ea0e67def10f8e4df9fafd1b6e983 [file] [log] [blame]
Ed Warnickecb9cada2015-12-08 15:45:58 -07001/*
2 * Copyright (c) 2015 Cisco and/or its affiliates.
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
6 *
7 * http://www.apache.org/licenses/LICENSE-2.0
8 *
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
14 */
15#include <vnet/vxlan/vxlan.h>
Chris Luke99cb3352016-04-26 10:49:53 -040016#include <vnet/ip/format.h>
John Loc42912d2016-11-07 18:30:47 -050017#include <vnet/fib/fib_entry.h>
18#include <vnet/fib/fib_table.h>
Neale Ranns1f50bf82019-07-16 15:28:52 +000019#include <vnet/fib/fib_entry_track.h>
Neale Ranns32e1c012016-11-22 17:07:28 +000020#include <vnet/mfib/mfib_table.h>
21#include <vnet/adj/adj_mcast.h>
eyal bari82e21d72018-04-26 13:14:55 +030022#include <vnet/adj/rewrite.h>
Neale Ranns1b5ca982020-12-16 13:06:58 +000023#include <vnet/dpo/drop_dpo.h>
Eyal Bari3212c572017-03-06 11:47:50 +020024#include <vnet/interface.h>
eyal bariaf86a482018-04-17 11:20:27 +030025#include <vnet/flow/flow.h>
Florin Corasb040f982020-10-20 14:59:43 -070026#include <vnet/udp/udp_local.h>
Hongjun Nibeb4bf72016-11-25 00:03:46 +080027#include <vlib/vlib.h>
Ed Warnickecb9cada2015-12-08 15:45:58 -070028
Billy McFallc5d9cda2016-09-14 10:39:58 -040029/**
30 * @file
31 * @brief VXLAN.
32 *
33 * VXLAN provides the features needed to allow L2 bridge domains (BDs)
34 * to span multiple servers. This is done by building an L2 overlay on
35 * top of an L3 network underlay using VXLAN tunnels.
36 *
37 * This makes it possible for servers to be co-located in the same data
38 * center or be separated geographically as long as they are reachable
39 * through the underlay L3 network.
40 *
41 * You can refer to this kind of L2 overlay bridge domain as a VXLAN
42 * (Virtual eXtensible VLAN) segment.
43 */
44
45
Ed Warnickecb9cada2015-12-08 15:45:58 -070046vxlan_main_t vxlan_main;
47
Ed Warnickea4b05412021-01-18 11:56:22 -060048static u32
49vxlan_eth_flag_change (vnet_main_t *vnm, vnet_hw_interface_t *hi, u32 flags)
50{
51 /* nothing for now */
52 return 0;
53}
54
Eyal Bari8f576482018-05-07 10:13:44 +030055static u8 *
56format_decap_next (u8 * s, va_list * args)
Hongjun Nibeb4bf72016-11-25 00:03:46 +080057{
58 u32 next_index = va_arg (*args, u32);
59
John Lo4478d8e2018-01-12 17:15:25 -050060 if (next_index == VXLAN_INPUT_NEXT_DROP)
61 return format (s, "drop");
62 else
63 return format (s, "index %d", next_index);
Hongjun Nibeb4bf72016-11-25 00:03:46 +080064 return s;
65}
66
Eyal Bari8f576482018-05-07 10:13:44 +030067u8 *
68format_vxlan_tunnel (u8 * s, va_list * args)
Ed Warnickecb9cada2015-12-08 15:45:58 -070069{
Eyal Bari8f576482018-05-07 10:13:44 +030070 vxlan_tunnel_t *t = va_arg (*args, vxlan_tunnel_t *);
Jon Loeliger3d460bd2018-02-01 16:36:12 -060071
72 s = format (s,
Artem Glazychev839dcc02020-12-01 02:39:21 +070073 "[%d] instance %d src %U dst %U src_port %d dst_port %d vni %d "
74 "fib-idx %d sw-if-idx %d ",
75 t->dev_instance, t->user_instance, format_ip46_address, &t->src,
76 IP46_TYPE_ANY, format_ip46_address, &t->dst, IP46_TYPE_ANY,
77 t->src_port, t->dst_port, t->vni, t->encap_fib_index,
78 t->sw_if_index);
John Lo56912c82016-12-08 16:10:02 -050079
John Lo4478d8e2018-01-12 17:15:25 -050080 s = format (s, "encap-dpo-idx %d ", t->next_dpo.dpoi_index);
John Lo56912c82016-12-08 16:10:02 -050081
John Lo4478d8e2018-01-12 17:15:25 -050082 if (PREDICT_FALSE (t->decap_next_index != VXLAN_INPUT_NEXT_L2_INPUT))
83 s = format (s, "decap-next-%U ", format_decap_next, t->decap_next_index);
84
Steven Luongb8de7d42021-12-13 11:26:30 -080085 s = format (s, "l3 %u ", t->is_l3);
86
John Lo4478d8e2018-01-12 17:15:25 -050087 if (PREDICT_FALSE (ip46_address_is_multicast (&t->dst)))
88 s = format (s, "mcast-sw-if-idx %d ", t->mcast_sw_if_index);
89
eyal bariaf86a482018-04-17 11:20:27 +030090 if (t->flow_index != ~0)
91 s = format (s, "flow-index %d [%U]", t->flow_index,
92 format_flow_enabled_hw, t->flow_index);
93
Ed Warnickecb9cada2015-12-08 15:45:58 -070094 return s;
95}
96
Eyal Bari8f576482018-05-07 10:13:44 +030097static u8 *
98format_vxlan_name (u8 * s, va_list * args)
Ed Warnickecb9cada2015-12-08 15:45:58 -070099{
100 u32 dev_instance = va_arg (*args, u32);
Eyal Bari8f576482018-05-07 10:13:44 +0300101 vxlan_main_t *vxm = &vxlan_main;
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600102 vxlan_tunnel_t *t;
103
104 if (dev_instance == ~0)
Eyal Bari8f576482018-05-07 10:13:44 +0300105 return format (s, "<cached-unused>");
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600106
Eyal Bari8f576482018-05-07 10:13:44 +0300107 if (dev_instance >= vec_len (vxm->tunnels))
108 return format (s, "<improperly-referenced>");
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600109
Eyal Bari8f576482018-05-07 10:13:44 +0300110 t = pool_elt_at_index (vxm->tunnels, dev_instance);
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600111
112 return format (s, "vxlan_tunnel%d", t->user_instance);
Ed Warnickecb9cada2015-12-08 15:45:58 -0700113}
114
Pavel Kotucek988a7c42016-02-29 15:03:08 +0100115static clib_error_t *
116vxlan_interface_admin_up_down (vnet_main_t * vnm, u32 hw_if_index, u32 flags)
117{
Eyal Baria93ea422017-02-01 13:36:15 +0200118 u32 hw_flags = (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP) ?
119 VNET_HW_INTERFACE_FLAG_LINK_UP : 0;
120 vnet_hw_interface_set_flags (vnm, hw_if_index, hw_flags);
Pavel Kotucek988a7c42016-02-29 15:03:08 +0100121
122 return /* no error */ 0;
123}
124
Eyal Bari8f576482018-05-07 10:13:44 +0300125/* *INDENT-OFF* */
126VNET_DEVICE_CLASS (vxlan_device_class, static) = {
Ed Warnickecb9cada2015-12-08 15:45:58 -0700127 .name = "VXLAN",
128 .format_device_name = format_vxlan_name,
129 .format_tx_trace = format_vxlan_encap_trace,
Pavel Kotucek988a7c42016-02-29 15:03:08 +0100130 .admin_up_down_function = vxlan_interface_admin_up_down,
Ed Warnickecb9cada2015-12-08 15:45:58 -0700131};
Eyal Bari8f576482018-05-07 10:13:44 +0300132/* *INDENT-ON* */
Ed Warnickecb9cada2015-12-08 15:45:58 -0700133
Eyal Bari8f576482018-05-07 10:13:44 +0300134static u8 *
135format_vxlan_header_with_length (u8 * s, va_list * args)
Ed Warnickecb9cada2015-12-08 15:45:58 -0700136{
137 u32 dev_instance = va_arg (*args, u32);
138 s = format (s, "unimplemented dev %u", dev_instance);
139 return s;
140}
141
Eyal Bari8f576482018-05-07 10:13:44 +0300142/* *INDENT-OFF* */
Ed Warnickecb9cada2015-12-08 15:45:58 -0700143VNET_HW_INTERFACE_CLASS (vxlan_hw_class) = {
144 .name = "VXLAN",
145 .format_header = format_vxlan_header_with_length,
Neale Rannsb80c5362016-10-08 13:03:40 +0100146 .build_rewrite = default_build_rewrite,
Ed Warnickecb9cada2015-12-08 15:45:58 -0700147};
Eyal Bari8f576482018-05-07 10:13:44 +0300148/* *INDENT-ON* */
Ed Warnickecb9cada2015-12-08 15:45:58 -0700149
Eyal Baric5b13602016-11-24 19:42:43 +0200150static void
Eyal Bari8f576482018-05-07 10:13:44 +0300151vxlan_tunnel_restack_dpo (vxlan_tunnel_t * t)
Eyal Baric5b13602016-11-24 19:42:43 +0200152{
Eyal Bari8f576482018-05-07 10:13:44 +0300153 u8 is_ip4 = ip46_address_is_ip4 (&t->dst);
Eyal Bari99ed4862018-04-25 13:50:57 +0300154 dpo_id_t dpo = DPO_INVALID;
155 fib_forward_chain_type_t forw_type = is_ip4 ?
Eyal Bari8f576482018-05-07 10:13:44 +0300156 FIB_FORW_CHAIN_TYPE_UNICAST_IP4 : FIB_FORW_CHAIN_TYPE_UNICAST_IP6;
Eyal Baric5b13602016-11-24 19:42:43 +0200157
Eyal Bari99ed4862018-04-25 13:50:57 +0300158 fib_entry_contribute_forwarding (t->fib_entry_index, forw_type, &dpo);
159
160 /* vxlan uses the payload hash as the udp source port
161 * hence the packet's hash is unknown
162 * skip single bucket load balance dpo's */
163 while (DPO_LOAD_BALANCE == dpo.dpoi_type)
164 {
Neale Ranns1b5ca982020-12-16 13:06:58 +0000165 const load_balance_t *lb;
166 const dpo_id_t *choice;
167
168 lb = load_balance_get (dpo.dpoi_index);
Eyal Bari99ed4862018-04-25 13:50:57 +0300169 if (lb->lb_n_buckets > 1)
Eyal Bari8f576482018-05-07 10:13:44 +0300170 break;
Eyal Bari99ed4862018-04-25 13:50:57 +0300171
Neale Ranns1b5ca982020-12-16 13:06:58 +0000172 choice = load_balance_get_bucket_i (lb, 0);
173
174 if (DPO_RECEIVE == choice->dpoi_type)
175 dpo_copy (&dpo, drop_dpo_get (choice->dpoi_proto));
176 else
177 dpo_copy (&dpo, choice);
Eyal Bari99ed4862018-04-25 13:50:57 +0300178 }
179
180 u32 encap_index = is_ip4 ?
Eyal Bari8f576482018-05-07 10:13:44 +0300181 vxlan4_encap_node.index : vxlan6_encap_node.index;
Eyal Bari99ed4862018-04-25 13:50:57 +0300182 dpo_stack_from_node (encap_index, &t->next_dpo, &dpo);
Eyal Bari8f576482018-05-07 10:13:44 +0300183 dpo_reset (&dpo);
Eyal Baric5b13602016-11-24 19:42:43 +0200184}
John Loc42912d2016-11-07 18:30:47 -0500185
186static vxlan_tunnel_t *
Eyal Bari8f576482018-05-07 10:13:44 +0300187vxlan_tunnel_from_fib_node (fib_node_t * node)
John Loc42912d2016-11-07 18:30:47 -0500188{
Eyal Bari8f576482018-05-07 10:13:44 +0300189 ASSERT (FIB_NODE_TYPE_VXLAN_TUNNEL == node->fn_type);
190 return ((vxlan_tunnel_t *) (((char *) node) -
191 STRUCT_OFFSET_OF (vxlan_tunnel_t, node)));
John Loc42912d2016-11-07 18:30:47 -0500192}
193
194/**
195 * Function definition to backwalk a FIB node -
196 * Here we will restack the new dpo of VXLAN DIP to encap node.
197 */
198static fib_node_back_walk_rc_t
Eyal Bari8f576482018-05-07 10:13:44 +0300199vxlan_tunnel_back_walk (fib_node_t * node, fib_node_back_walk_ctx_t * ctx)
John Loc42912d2016-11-07 18:30:47 -0500200{
Eyal Bari8f576482018-05-07 10:13:44 +0300201 vxlan_tunnel_restack_dpo (vxlan_tunnel_from_fib_node (node));
202 return (FIB_NODE_BACK_WALK_CONTINUE);
John Loc42912d2016-11-07 18:30:47 -0500203}
204
205/**
206 * Function definition to get a FIB node from its index
207 */
Eyal Bari8f576482018-05-07 10:13:44 +0300208static fib_node_t *
John Loc42912d2016-11-07 18:30:47 -0500209vxlan_tunnel_fib_node_get (fib_node_index_t index)
210{
Eyal Bari8f576482018-05-07 10:13:44 +0300211 vxlan_tunnel_t *t;
212 vxlan_main_t *vxm = &vxlan_main;
John Loc42912d2016-11-07 18:30:47 -0500213
Eyal Bari8f576482018-05-07 10:13:44 +0300214 t = pool_elt_at_index (vxm->tunnels, index);
John Loc42912d2016-11-07 18:30:47 -0500215
Eyal Bari8f576482018-05-07 10:13:44 +0300216 return (&t->node);
John Loc42912d2016-11-07 18:30:47 -0500217}
218
219/**
220 * Function definition to inform the FIB node that its last lock has gone.
221 */
222static void
Eyal Bari8f576482018-05-07 10:13:44 +0300223vxlan_tunnel_last_lock_gone (fib_node_t * node)
John Loc42912d2016-11-07 18:30:47 -0500224{
Eyal Bari8f576482018-05-07 10:13:44 +0300225 /*
226 * The VXLAN tunnel is a root of the graph. As such
227 * it never has children and thus is never locked.
228 */
229 ASSERT (0);
John Loc42912d2016-11-07 18:30:47 -0500230}
231
232/*
233 * Virtual function table registered by VXLAN tunnels
234 * for participation in the FIB object graph.
235 */
236const static fib_node_vft_t vxlan_vft = {
Eyal Bari8f576482018-05-07 10:13:44 +0300237 .fnv_get = vxlan_tunnel_fib_node_get,
238 .fnv_last_lock = vxlan_tunnel_last_lock_gone,
239 .fnv_back_walk = vxlan_tunnel_back_walk,
John Loc42912d2016-11-07 18:30:47 -0500240};
241
Artem Glazychev839dcc02020-12-01 02:39:21 +0700242#define foreach_copy_field \
243 _ (vni) \
244 _ (mcast_sw_if_index) \
245 _ (encap_fib_index) \
246 _ (decap_next_index) \
247 _ (src) \
248 _ (dst) \
249 _ (src_port) \
Steven Luongb8de7d42021-12-13 11:26:30 -0800250 _ (dst_port) \
251 _ (is_l3)
Chris Luke99cb3352016-04-26 10:49:53 -0400252
Eyal Bari554075a2018-02-13 15:17:17 +0200253static void
Eyal Baria93ea422017-02-01 13:36:15 +0200254vxlan_rewrite (vxlan_tunnel_t * t, bool is_ip6)
Ed Warnickecb9cada2015-12-08 15:45:58 -0700255{
Eyal Bari8f576482018-05-07 10:13:44 +0300256 union
257 {
eyal bari82e21d72018-04-26 13:14:55 +0300258 ip4_vxlan_header_t h4;
259 ip6_vxlan_header_t h6;
Matthew Smithec8aea12018-05-02 09:55:01 -0500260 } h;
eyal bari82e21d72018-04-26 13:14:55 +0300261 int len = is_ip6 ? sizeof h.h6 : sizeof h.h4;
Ed Warnickecb9cada2015-12-08 15:45:58 -0700262
Eyal Bari8f576482018-05-07 10:13:44 +0300263 udp_header_t *udp;
264 vxlan_header_t *vxlan;
Eyal Baria93ea422017-02-01 13:36:15 +0200265 /* Fixed portion of the (outer) ip header */
Matthew Smithec8aea12018-05-02 09:55:01 -0500266
Dave Barachb7b92992018-10-17 10:38:51 -0400267 clib_memset (&h, 0, sizeof (h));
Eyal Bari8f576482018-05-07 10:13:44 +0300268 if (!is_ip6)
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800269 {
Eyal Bari8f576482018-05-07 10:13:44 +0300270 ip4_header_t *ip = &h.h4.ip4;
eyal bari82e21d72018-04-26 13:14:55 +0300271 udp = &h.h4.udp, vxlan = &h.h4.vxlan;
Eyal Baria93ea422017-02-01 13:36:15 +0200272 ip->ip_version_and_header_length = 0x45;
273 ip->ttl = 254;
274 ip->protocol = IP_PROTOCOL_UDP;
Eyal Bari8f576482018-05-07 10:13:44 +0300275
Eyal Baria93ea422017-02-01 13:36:15 +0200276 ip->src_address = t->src.ip4;
277 ip->dst_address = t->dst.ip4;
278
279 /* we fix up the ip4 header length and checksum after-the-fact */
280 ip->checksum = ip4_header_checksum (ip);
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800281 }
282 else
283 {
Eyal Bari8f576482018-05-07 10:13:44 +0300284 ip6_header_t *ip = &h.h6.ip6;
eyal bari82e21d72018-04-26 13:14:55 +0300285 udp = &h.h6.udp, vxlan = &h.h6.vxlan;
Eyal Bari8f576482018-05-07 10:13:44 +0300286 ip->ip_version_traffic_class_and_flow_label =
287 clib_host_to_net_u32 (6 << 28);
Eyal Baria93ea422017-02-01 13:36:15 +0200288 ip->hop_limit = 255;
289 ip->protocol = IP_PROTOCOL_UDP;
Eyal Bari8f576482018-05-07 10:13:44 +0300290
Eyal Baria93ea422017-02-01 13:36:15 +0200291 ip->src_address = t->src.ip6;
292 ip->dst_address = t->dst.ip6;
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800293 }
294
Eyal Baria93ea422017-02-01 13:36:15 +0200295 /* UDP header, randomize src port on something, maybe? */
Artem Glazychev839dcc02020-12-01 02:39:21 +0700296 udp->src_port = clib_host_to_net_u16 (t->src_port);
297 udp->dst_port = clib_host_to_net_u16 (t->dst_port);
Eyal Baria93ea422017-02-01 13:36:15 +0200298
299 /* VXLAN header */
Eyal Bari8f576482018-05-07 10:13:44 +0300300 vnet_set_vni_and_flags (vxlan, t->vni);
eyal bari82e21d72018-04-26 13:14:55 +0300301 vnet_rewrite_set_data (*t, &h, len);
Eyal Baria93ea422017-02-01 13:36:15 +0200302}
303
304static bool
Eyal Bari8f576482018-05-07 10:13:44 +0300305vxlan_decap_next_is_valid (vxlan_main_t * vxm, u32 is_ip6,
306 u32 decap_next_index)
Eyal Baria93ea422017-02-01 13:36:15 +0200307{
Eyal Bari8f576482018-05-07 10:13:44 +0300308 vlib_main_t *vm = vxm->vlib_main;
309 u32 input_idx = (!is_ip6) ?
310 vxlan4_input_node.index : vxlan6_input_node.index;
Eyal Baria93ea422017-02-01 13:36:15 +0200311 vlib_node_runtime_t *r = vlib_node_get_runtime (vm, input_idx);
312
313 return decap_next_index < r->n_next_nodes;
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800314}
315
Eyal Bari8f576482018-05-07 10:13:44 +0300316/* *INDENT-OFF* */
317typedef CLIB_PACKED(union
318{
319 struct
320 {
Neale Ranns32e1c012016-11-22 17:07:28 +0000321 fib_node_index_t mfib_entry_index;
Eyal Bari0ded8512017-01-19 17:01:09 +0200322 adj_index_t mcast_adj_index;
323 };
324 u64 as_u64;
325}) mcast_shared_t;
Eyal Bari8f576482018-05-07 10:13:44 +0300326/* *INDENT-ON* */
Eyal Bari0ded8512017-01-19 17:01:09 +0200327
328static inline mcast_shared_t
Eyal Bari8f576482018-05-07 10:13:44 +0300329mcast_shared_get (ip46_address_t * ip)
Eyal Barifff67c82016-12-21 12:45:47 +0200330{
Eyal Bari8f576482018-05-07 10:13:44 +0300331 ASSERT (ip46_address_is_multicast (ip));
332 uword *p = hash_get_mem (vxlan_main.mcast_shared, ip);
Dave Barach47d41ad2020-02-17 09:13:26 -0500333 ALWAYS_ASSERT (p);
Eyal Bari8f576482018-05-07 10:13:44 +0300334 mcast_shared_t ret = {.as_u64 = *p };
335 return ret;
Eyal Barifff67c82016-12-21 12:45:47 +0200336}
337
Eyal Bari0ded8512017-01-19 17:01:09 +0200338static inline void
Eyal Bari8f576482018-05-07 10:13:44 +0300339mcast_shared_add (ip46_address_t * dst, fib_node_index_t mfei, adj_index_t ai)
Neale Ranns32e1c012016-11-22 17:07:28 +0000340{
Eyal Bari8f576482018-05-07 10:13:44 +0300341 mcast_shared_t new_ep = {
342 .mcast_adj_index = ai,
343 .mfib_entry_index = mfei,
344 };
Neale Ranns32e1c012016-11-22 17:07:28 +0000345
Eyal Bari8f576482018-05-07 10:13:44 +0300346 hash_set_mem_alloc (&vxlan_main.mcast_shared, dst, new_ep.as_u64);
Neale Ranns32e1c012016-11-22 17:07:28 +0000347}
348
349static inline void
Eyal Bari8f576482018-05-07 10:13:44 +0300350mcast_shared_remove (ip46_address_t * dst)
Neale Ranns32e1c012016-11-22 17:07:28 +0000351{
Eyal Bari8f576482018-05-07 10:13:44 +0300352 mcast_shared_t ep = mcast_shared_get (dst);
Neale Ranns32e1c012016-11-22 17:07:28 +0000353
Eyal Bari8f576482018-05-07 10:13:44 +0300354 adj_unlock (ep.mcast_adj_index);
355 mfib_table_entry_delete_index (ep.mfib_entry_index, MFIB_SOURCE_VXLAN);
Neale Ranns32e1c012016-11-22 17:07:28 +0000356
Eyal Bari8f576482018-05-07 10:13:44 +0300357 hash_unset_mem_free (&vxlan_main.mcast_shared, dst);
Eyal Barifff67c82016-12-21 12:45:47 +0200358}
359
Eyal Bari8f576482018-05-07 10:13:44 +0300360int vnet_vxlan_add_del_tunnel
361 (vnet_vxlan_add_del_tunnel_args_t * a, u32 * sw_if_indexp)
Ed Warnickecb9cada2015-12-08 15:45:58 -0700362{
Eyal Bari8f576482018-05-07 10:13:44 +0300363 vxlan_main_t *vxm = &vxlan_main;
Eyal Bari8f576482018-05-07 10:13:44 +0300364 vnet_main_t *vnm = vxm->vnet_main;
Eyal Bariefd9cf32018-10-02 12:23:06 +0300365 vxlan_decap_info_t *p;
Ed Warnickecb9cada2015-12-08 15:45:58 -0700366 u32 sw_if_index = ~0;
Chris Luke99cb3352016-04-26 10:49:53 -0400367 vxlan4_tunnel_key_t key4;
368 vxlan6_tunnel_key_t key6;
John Loc42912d2016-11-07 18:30:47 -0500369 u32 is_ip6 = a->is_ip6;
Ed Warnickea4b05412021-01-18 11:56:22 -0600370 vlib_main_t *vm = vlib_get_main ();
371 u8 hw_addr[6];
Ed Warnickecb9cada2015-12-08 15:45:58 -0700372
Artem Glazychev839dcc02020-12-01 02:39:21 +0700373 /* Set udp-ports */
374 if (a->src_port == 0)
375 a->src_port = is_ip6 ? UDP_DST_PORT_vxlan6 : UDP_DST_PORT_vxlan;
376
377 if (a->dst_port == 0)
378 a->dst_port = is_ip6 ? UDP_DST_PORT_vxlan6 : UDP_DST_PORT_vxlan;
379
Eyal Baridd47eca2018-07-08 08:15:56 +0300380 int not_found;
John Lo56912c82016-12-08 16:10:02 -0500381 if (!is_ip6)
382 {
Eyal Bariefd9cf32018-10-02 12:23:06 +0300383 /* ip4 mcast is indexed by mcast addr only */
384 key4.key[0] = ip46_address_is_multicast (&a->dst) ?
Artem Glazychev839dcc02020-12-01 02:39:21 +0700385 a->dst.ip4.as_u32 :
386 a->dst.ip4.as_u32 | (((u64) a->src.ip4.as_u32) << 32);
387 key4.key[1] = ((u64) clib_host_to_net_u16 (a->src_port) << 48) |
388 (((u64) a->encap_fib_index) << 32) |
389 clib_host_to_net_u32 (a->vni << 8);
Eyal Baridd47eca2018-07-08 08:15:56 +0300390 not_found =
391 clib_bihash_search_inline_16_8 (&vxm->vxlan4_tunnel_by_key, &key4);
Eyal Bariefd9cf32018-10-02 12:23:06 +0300392 p = (void *) &key4.value;
Eyal Bari8f576482018-05-07 10:13:44 +0300393 }
394 else
John Lo56912c82016-12-08 16:10:02 -0500395 {
Eyal Bari0fa56782018-06-04 12:25:05 +0300396 key6.key[0] = a->dst.ip6.as_u64[0];
397 key6.key[1] = a->dst.ip6.as_u64[1];
Artem Glazychev839dcc02020-12-01 02:39:21 +0700398 key6.key[2] = (((u64) clib_host_to_net_u16 (a->src_port) << 48) |
399 ((u64) a->encap_fib_index) << 32) |
400 clib_host_to_net_u32 (a->vni << 8);
Eyal Baridd47eca2018-07-08 08:15:56 +0300401 not_found =
402 clib_bihash_search_inline_24_8 (&vxm->vxlan6_tunnel_by_key, &key6);
Eyal Bariefd9cf32018-10-02 12:23:06 +0300403 p = (void *) &key6.value;
John Lo56912c82016-12-08 16:10:02 -0500404 }
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600405
Eyal Baridd47eca2018-07-08 08:15:56 +0300406 if (not_found)
407 p = 0;
408
Ed Warnickecb9cada2015-12-08 15:45:58 -0700409 if (a->is_add)
410 {
Eyal Bari8f576482018-05-07 10:13:44 +0300411 l2input_main_t *l2im = &l2input_main;
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600412 u32 dev_instance; /* real dev instance tunnel index */
413 u32 user_instance; /* request and actual instance number */
John Loc42912d2016-11-07 18:30:47 -0500414
Ed Warnickecb9cada2015-12-08 15:45:58 -0700415 /* adding a tunnel: tunnel must not already exist */
416 if (p)
Eyal Bari8f576482018-05-07 10:13:44 +0300417 return VNET_API_ERROR_TUNNEL_EXIST;
Ed Warnickecb9cada2015-12-08 15:45:58 -0700418
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800419 /*if not set explicitly, default to l2 */
Eyal Bari8f576482018-05-07 10:13:44 +0300420 if (a->decap_next_index == ~0)
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800421 a->decap_next_index = VXLAN_INPUT_NEXT_L2_INPUT;
Eyal Bari8f576482018-05-07 10:13:44 +0300422 if (!vxlan_decap_next_is_valid (vxm, is_ip6, a->decap_next_index))
423 return VNET_API_ERROR_INVALID_DECAP_NEXT;
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800424
Eyal Bariefd9cf32018-10-02 12:23:06 +0300425 vxlan_tunnel_t *t;
Ed Warnickecb9cada2015-12-08 15:45:58 -0700426 pool_get_aligned (vxm->tunnels, t, CLIB_CACHE_LINE_BYTES);
Dave Barachb7b92992018-10-17 10:38:51 -0400427 clib_memset (t, 0, sizeof (*t));
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600428 dev_instance = t - vxm->tunnels;
429
Ed Warnickecb9cada2015-12-08 15:45:58 -0700430 /* copy from arg structure */
431#define _(x) t->x = a->x;
432 foreach_copy_field;
433#undef _
Chris Luke99cb3352016-04-26 10:49:53 -0400434
Eyal Bari554075a2018-02-13 15:17:17 +0200435 vxlan_rewrite (t, is_ip6);
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600436 /*
437 * Reconcile the real dev_instance and a possible requested instance.
438 */
439 user_instance = a->instance;
Eyal Bari8f576482018-05-07 10:13:44 +0300440 if (user_instance == ~0)
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600441 user_instance = dev_instance;
442 if (hash_get (vxm->instance_used, user_instance))
443 {
444 pool_put (vxm->tunnels, t);
445 return VNET_API_ERROR_INSTANCE_IN_USE;
446 }
Ed Warnickea4b05412021-01-18 11:56:22 -0600447
Ray Kinsella15cb3da2021-01-28 17:09:45 +0000448 hash_set (vxm->instance_used, user_instance, 1);
449
450 t->dev_instance = dev_instance; /* actual */
451 t->user_instance = user_instance; /* name */
452 t->flow_index = ~0;
453
Steven Luongb8de7d42021-12-13 11:26:30 -0800454 if (a->is_l3 == 0)
Nathan Skrzypczak3e384222021-04-21 19:56:34 +0200455 t->hw_if_index =
456 vnet_register_interface (vnm, vxlan_device_class.index, dev_instance,
457 vxlan_hw_class.index, dev_instance);
458 else
Ed Warnickea4b05412021-01-18 11:56:22 -0600459 {
Nathan Skrzypczak3e384222021-04-21 19:56:34 +0200460 f64 now = vlib_time_now (vm);
461 u32 rnd;
462 rnd = (u32) (now * 1e6);
463 rnd = random_u32 (&rnd);
464 memcpy (hw_addr + 2, &rnd, sizeof (rnd));
465 hw_addr[0] = 2;
466 hw_addr[1] = 0xfe;
467 if (ethernet_register_interface (
468 vnm, vxlan_device_class.index, dev_instance, hw_addr,
469 &t->hw_if_index, vxlan_eth_flag_change))
470 {
471 hash_unset (vxm->instance_used, t->user_instance);
Ray Kinsella15cb3da2021-01-28 17:09:45 +0000472
Nathan Skrzypczak3e384222021-04-21 19:56:34 +0200473 pool_put (vxm->tunnels, t);
474 return VNET_API_ERROR_SYSCALL_ERROR_2;
475 }
Ed Warnickea4b05412021-01-18 11:56:22 -0600476 }
477
Eyal Bari8f576482018-05-07 10:13:44 +0300478 vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, t->hw_if_index);
John Loe5453d02018-01-23 19:21:34 -0500479
480 /* Set vxlan tunnel output node */
481 u32 encap_index = !is_ip6 ?
Eyal Bari8f576482018-05-07 10:13:44 +0300482 vxlan4_encap_node.index : vxlan6_encap_node.index;
Eyal Bari554075a2018-02-13 15:17:17 +0200483 vnet_set_interface_output_node (vnm, t->hw_if_index, encap_index);
John Loe5453d02018-01-23 19:21:34 -0500484
Ed Warnickecb9cada2015-12-08 15:45:58 -0700485 t->sw_if_index = sw_if_index = hi->sw_if_index;
John Loe5453d02018-01-23 19:21:34 -0500486
Eyal Bariefd9cf32018-10-02 12:23:06 +0300487 /* copy the key */
488 int add_failed;
489 if (is_ip6)
490 {
491 key6.value = (u64) dev_instance;
492 add_failed = clib_bihash_add_del_24_8 (&vxm->vxlan6_tunnel_by_key,
493 &key6, 1 /*add */ );
494 }
495 else
496 {
497 vxlan_decap_info_t di = {.sw_if_index = t->sw_if_index, };
498 if (ip46_address_is_multicast (&t->dst))
499 di.local_ip = t->src.ip4;
500 else
501 di.next_index = t->decap_next_index;
502 key4.value = di.as_u64;
503 add_failed = clib_bihash_add_del_16_8 (&vxm->vxlan4_tunnel_by_key,
504 &key4, 1 /*add */ );
505 }
506
507 if (add_failed)
508 {
Steven Luongb8de7d42021-12-13 11:26:30 -0800509 if (a->is_l3 == 0)
Nathan Skrzypczak3e384222021-04-21 19:56:34 +0200510 vnet_delete_hw_interface (vnm, t->hw_if_index);
511 else
512 ethernet_delete_interface (vnm, t->hw_if_index);
Eyal Bariefd9cf32018-10-02 12:23:06 +0300513 hash_unset (vxm->instance_used, t->user_instance);
514 pool_put (vxm->tunnels, t);
515 return VNET_API_ERROR_INVALID_REGISTRATION;
516 }
517
Eyal Bari8f576482018-05-07 10:13:44 +0300518 vec_validate_init_empty (vxm->tunnel_index_by_sw_if_index, sw_if_index,
519 ~0);
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600520 vxm->tunnel_index_by_sw_if_index[sw_if_index] = dev_instance;
Dave Wallace60231f32015-12-17 21:04:30 -0500521
John Loc42912d2016-11-07 18:30:47 -0500522 /* setup l2 input config with l2 feature and bd 0 to drop packet */
523 vec_validate (l2im->configs, sw_if_index);
524 l2im->configs[sw_if_index].feature_bitmap = L2INPUT_FEAT_DROP;
525 l2im->configs[sw_if_index].bd_index = 0;
John Loe5453d02018-01-23 19:21:34 -0500526
Eyal Bari8f576482018-05-07 10:13:44 +0300527 vnet_sw_interface_t *si = vnet_get_sw_interface (vnm, sw_if_index);
Eyal Bari3212c572017-03-06 11:47:50 +0200528 si->flags &= ~VNET_SW_INTERFACE_FLAG_HIDDEN;
John Loe5453d02018-01-23 19:21:34 -0500529 vnet_sw_interface_set_flags (vnm, sw_if_index,
Eyal Bari8f576482018-05-07 10:13:44 +0300530 VNET_SW_INTERFACE_FLAG_ADMIN_UP);
Eyal Bari3212c572017-03-06 11:47:50 +0200531
Eyal Bari8f576482018-05-07 10:13:44 +0300532 fib_node_init (&t->node, FIB_NODE_TYPE_VXLAN_TUNNEL);
Eyal Baric5b13602016-11-24 19:42:43 +0200533 fib_prefix_t tun_dst_pfx;
Eyal Baric5b13602016-11-24 19:42:43 +0200534 vnet_flood_class_t flood_class = VNET_FLOOD_CLASS_TUNNEL_NORMAL;
John Loc42912d2016-11-07 18:30:47 -0500535
Eyal Bari8f576482018-05-07 10:13:44 +0300536 fib_prefix_from_ip46_addr (&t->dst, &tun_dst_pfx);
537 if (!ip46_address_is_multicast (&t->dst))
538 {
539 /* Unicast tunnel -
540 * source the FIB entry for the tunnel's destination
541 * and become a child thereof. The tunnel will then get poked
542 * when the forwarding for the entry updates, and the tunnel can
543 * re-stack accordingly
544 */
Nick Zavaritsky27518c22020-02-27 15:54:58 +0000545 vtep_addr_ref (&vxm->vtep_table, t->encap_fib_index, &t->src);
Neale Ranns1f50bf82019-07-16 15:28:52 +0000546 t->fib_entry_index = fib_entry_track (t->encap_fib_index,
547 &tun_dst_pfx,
548 FIB_NODE_TYPE_VXLAN_TUNNEL,
549 dev_instance,
550 &t->sibling_index);
Eyal Bari8f576482018-05-07 10:13:44 +0300551 vxlan_tunnel_restack_dpo (t);
552 }
Eyal Barifff67c82016-12-21 12:45:47 +0200553 else
Eyal Bari8f576482018-05-07 10:13:44 +0300554 {
555 /* Multicast tunnel -
Paul Vinciguerrabdc0e6b2018-09-22 05:32:50 -0700556 * as the same mcast group can be used for multiple mcast tunnels
557 * with different VNIs, create the output fib adjacency only if
John Lo56912c82016-12-08 16:10:02 -0500558 * it does not already exist
559 */
Eyal Bari8f576482018-05-07 10:13:44 +0300560 fib_protocol_t fp = fib_ip_proto (is_ip6);
Eyal Baric5b13602016-11-24 19:42:43 +0200561
Nick Zavaritsky27518c22020-02-27 15:54:58 +0000562 if (vtep_addr_ref (&vxm->vtep_table,
563 t->encap_fib_index, &t->dst) == 1)
Eyal Bari8f576482018-05-07 10:13:44 +0300564 {
565 fib_node_index_t mfei;
566 adj_index_t ai;
567 fib_route_path_t path = {
568 .frp_proto = fib_proto_to_dpo (fp),
569 .frp_addr = zero_addr,
570 .frp_sw_if_index = 0xffffffff,
571 .frp_fib_index = ~0,
Neale Ranns097fa662018-05-01 05:17:55 -0700572 .frp_weight = 1,
Eyal Bari8f576482018-05-07 10:13:44 +0300573 .frp_flags = FIB_ROUTE_PATH_LOCAL,
Neale Ranns097fa662018-05-01 05:17:55 -0700574 .frp_mitf_flags = MFIB_ITF_FLAG_FORWARD,
Eyal Bari8f576482018-05-07 10:13:44 +0300575 };
576 const mfib_prefix_t mpfx = {
577 .fp_proto = fp,
578 .fp_len = (is_ip6 ? 128 : 32),
579 .fp_grp_addr = tun_dst_pfx.fp_addr,
580 };
Eyal Barifff67c82016-12-21 12:45:47 +0200581
Eyal Bari8f576482018-05-07 10:13:44 +0300582 /*
583 * Setup the (*,G) to receive traffic on the mcast group
584 * - the forwarding interface is for-us
585 * - the accepting interface is that from the API
586 */
Paul Atkins8e2b1b12021-10-12 14:32:11 +0100587 mfib_table_entry_path_update (t->encap_fib_index, &mpfx,
588 MFIB_SOURCE_VXLAN,
589 MFIB_ENTRY_FLAG_NONE, &path);
Neale Ranns32e1c012016-11-22 17:07:28 +0000590
Eyal Bari8f576482018-05-07 10:13:44 +0300591 path.frp_sw_if_index = a->mcast_sw_if_index;
592 path.frp_flags = FIB_ROUTE_PATH_FLAG_NONE;
Neale Ranns097fa662018-05-01 05:17:55 -0700593 path.frp_mitf_flags = MFIB_ITF_FLAG_ACCEPT;
Paul Atkins8e2b1b12021-10-12 14:32:11 +0100594 mfei = mfib_table_entry_path_update (
595 t->encap_fib_index, &mpfx, MFIB_SOURCE_VXLAN,
596 MFIB_ENTRY_FLAG_NONE, &path);
Neale Ranns32e1c012016-11-22 17:07:28 +0000597
Eyal Bari8f576482018-05-07 10:13:44 +0300598 /*
599 * Create the mcast adjacency to send traffic to the group
600 */
601 ai = adj_mcast_add_or_lock (fp,
602 fib_proto_to_link (fp),
603 a->mcast_sw_if_index);
Neale Ranns32e1c012016-11-22 17:07:28 +0000604
Eyal Bari8f576482018-05-07 10:13:44 +0300605 /*
606 * create a new end-point
607 */
608 mcast_shared_add (&t->dst, mfei, ai);
609 }
Neale Ranns32e1c012016-11-22 17:07:28 +0000610
Eyal Bari8f576482018-05-07 10:13:44 +0300611 dpo_id_t dpo = DPO_INVALID;
612 mcast_shared_t ep = mcast_shared_get (&t->dst);
Neale Ranns32e1c012016-11-22 17:07:28 +0000613
Eyal Bari8f576482018-05-07 10:13:44 +0300614 /* Stack shared mcast dst mac addr rewrite on encap */
615 dpo_set (&dpo, DPO_ADJACENCY_MCAST,
616 fib_proto_to_dpo (fp), ep.mcast_adj_index);
Neale Ranns32e1c012016-11-22 17:07:28 +0000617
Eyal Bari8f576482018-05-07 10:13:44 +0300618 dpo_stack_from_node (encap_index, &t->next_dpo, &dpo);
619 dpo_reset (&dpo);
Eyal Barifff67c82016-12-21 12:45:47 +0200620 flood_class = VNET_FLOOD_CLASS_TUNNEL_MASTER;
John Lo56912c82016-12-08 16:10:02 -0500621 }
622
Eyal Bari8f576482018-05-07 10:13:44 +0300623 vnet_get_sw_interface (vnet_get_main (), sw_if_index)->flood_class =
624 flood_class;
Ed Warnickecb9cada2015-12-08 15:45:58 -0700625 }
626 else
627 {
628 /* deleting a tunnel: tunnel must exist */
John Lo37682e12016-11-30 12:51:39 -0500629 if (!p)
Eyal Bari8f576482018-05-07 10:13:44 +0300630 return VNET_API_ERROR_NO_SUCH_ENTRY;
Ed Warnickecb9cada2015-12-08 15:45:58 -0700631
Eyal Barif8d5e212018-10-14 10:54:32 +0300632 u32 instance = is_ip6 ? key6.value :
633 vxm->tunnel_index_by_sw_if_index[p->sw_if_index];
Eyal Bariefd9cf32018-10-02 12:23:06 +0300634 vxlan_tunnel_t *t = pool_elt_at_index (vxm->tunnels, instance);
Ed Warnickecb9cada2015-12-08 15:45:58 -0700635
Eyal Baribc799c92017-04-06 03:26:59 +0300636 sw_if_index = t->sw_if_index;
Eyal Bari8f576482018-05-07 10:13:44 +0300637 vnet_sw_interface_set_flags (vnm, sw_if_index, 0 /* down */ );
Ed Warnickecb9cada2015-12-08 15:45:58 -0700638
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600639 vxm->tunnel_index_by_sw_if_index[sw_if_index] = ~0;
Dave Wallace60231f32015-12-17 21:04:30 -0500640
John Loc42912d2016-11-07 18:30:47 -0500641 if (!is_ip6)
Eyal Baridd47eca2018-07-08 08:15:56 +0300642 clib_bihash_add_del_16_8 (&vxm->vxlan4_tunnel_by_key, &key4,
643 0 /*del */ );
Chris Luke99cb3352016-04-26 10:49:53 -0400644 else
Eyal Baridd47eca2018-07-08 08:15:56 +0300645 clib_bihash_add_del_24_8 (&vxm->vxlan6_tunnel_by_key, &key6,
Eyal Bari0fa56782018-06-04 12:25:05 +0300646 0 /*del */ );
Eyal Barifff67c82016-12-21 12:45:47 +0200647
Eyal Bari8f576482018-05-07 10:13:44 +0300648 if (!ip46_address_is_multicast (&t->dst))
649 {
eyal bariaf86a482018-04-17 11:20:27 +0300650 if (t->flow_index != ~0)
651 vnet_flow_del (vnm, t->flow_index);
652
Nick Zavaritsky27518c22020-02-27 15:54:58 +0000653 vtep_addr_unref (&vxm->vtep_table, t->encap_fib_index, &t->src);
Neale Ranns1f50bf82019-07-16 15:28:52 +0000654 fib_entry_untrack (t->fib_entry_index, t->sibling_index);
Eyal Bari8f576482018-05-07 10:13:44 +0300655 }
Nick Zavaritsky27518c22020-02-27 15:54:58 +0000656 else if (vtep_addr_unref (&vxm->vtep_table,
657 t->encap_fib_index, &t->dst) == 0)
Eyal Bari8f576482018-05-07 10:13:44 +0300658 {
659 mcast_shared_remove (&t->dst);
660 }
Eyal Barifff67c82016-12-21 12:45:47 +0200661
Steven Luongb8de7d42021-12-13 11:26:30 -0800662 if (t->is_l3 == 0)
Nathan Skrzypczak3e384222021-04-21 19:56:34 +0200663 vnet_delete_hw_interface (vnm, t->hw_if_index);
664 else
665 ethernet_delete_interface (vnm, t->hw_if_index);
Jon Loeliger25ef2b52018-02-23 17:02:41 -0600666 hash_unset (vxm->instance_used, t->user_instance);
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600667
Eyal Bari8f576482018-05-07 10:13:44 +0300668 fib_node_deinit (&t->node);
Ed Warnickecb9cada2015-12-08 15:45:58 -0700669 pool_put (vxm->tunnels, t);
670 }
671
672 if (sw_if_indexp)
Eyal Bari8f576482018-05-07 10:13:44 +0300673 *sw_if_indexp = sw_if_index;
Ed Warnickecb9cada2015-12-08 15:45:58 -0700674
Jakub Grajciardf3ca232019-06-04 13:16:42 +0200675 if (a->is_add)
676 {
677 /* register udp ports */
Artem Glazychev839dcc02020-12-01 02:39:21 +0700678 if (!is_ip6 && !udp_is_valid_dst_port (a->src_port, 1))
679 udp_register_dst_port (vxm->vlib_main, a->src_port,
Jakub Grajciardf3ca232019-06-04 13:16:42 +0200680 vxlan4_input_node.index, 1);
Artem Glazychev839dcc02020-12-01 02:39:21 +0700681 if (is_ip6 && !udp_is_valid_dst_port (a->src_port, 0))
682 udp_register_dst_port (vxm->vlib_main, a->src_port,
Jakub Grajciardf3ca232019-06-04 13:16:42 +0200683 vxlan6_input_node.index, 0);
684 }
685
Ed Warnickecb9cada2015-12-08 15:45:58 -0700686 return 0;
687}
688
Eyal Bari8f576482018-05-07 10:13:44 +0300689static uword
690get_decap_next_for_node (u32 node_index, u32 ipv4_set)
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800691{
Eyal Bari8f576482018-05-07 10:13:44 +0300692 vxlan_main_t *vxm = &vxlan_main;
693 vlib_main_t *vm = vxm->vlib_main;
Eyal Bari0765c6e2017-01-29 12:40:50 +0200694 uword input_node = (ipv4_set) ? vxlan4_input_node.index :
695 vxlan6_input_node.index;
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800696
Eyal Bari0765c6e2017-01-29 12:40:50 +0200697 return vlib_node_add_next (vm, input_node, node_index);
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800698}
699
Eyal Bari8f576482018-05-07 10:13:44 +0300700static uword
701unformat_decap_next (unformat_input_t * input, va_list * args)
Ed Warnickecb9cada2015-12-08 15:45:58 -0700702{
Eyal Bari8f576482018-05-07 10:13:44 +0300703 u32 *result = va_arg (*args, u32 *);
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800704 u32 ipv4_set = va_arg (*args, int);
Eyal Bari8f576482018-05-07 10:13:44 +0300705 vxlan_main_t *vxm = &vxlan_main;
706 vlib_main_t *vm = vxm->vlib_main;
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800707 u32 node_index;
Ed Warnickecb9cada2015-12-08 15:45:58 -0700708 u32 tmp;
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800709
Ed Warnickecb9cada2015-12-08 15:45:58 -0700710 if (unformat (input, "l2"))
711 *result = VXLAN_INPUT_NEXT_L2_INPUT;
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800712 else if (unformat (input, "node %U", unformat_vlib_node, vm, &node_index))
Eyal Bari8f576482018-05-07 10:13:44 +0300713 *result = get_decap_next_for_node (node_index, ipv4_set);
Ed Warnickecb9cada2015-12-08 15:45:58 -0700714 else if (unformat (input, "%d", &tmp))
715 *result = tmp;
716 else
717 return 0;
718 return 1;
719}
720
721static clib_error_t *
722vxlan_add_del_tunnel_command_fn (vlib_main_t * vm,
Eyal Bari8f576482018-05-07 10:13:44 +0300723 unformat_input_t * input,
724 vlib_cli_command_t * cmd)
Ed Warnickecb9cada2015-12-08 15:45:58 -0700725{
Eyal Bari8f576482018-05-07 10:13:44 +0300726 unformat_input_t _line_input, *line_input = &_line_input;
727 ip46_address_t src = ip46_address_initializer, dst =
728 ip46_address_initializer;
Ed Warnickecb9cada2015-12-08 15:45:58 -0700729 u8 is_add = 1;
730 u8 src_set = 0;
731 u8 dst_set = 0;
Eyal Baric5b13602016-11-24 19:42:43 +0200732 u8 grp_set = 0;
Chris Luke99cb3352016-04-26 10:49:53 -0400733 u8 ipv4_set = 0;
734 u8 ipv6_set = 0;
Nathan Skrzypczak3e384222021-04-21 19:56:34 +0200735 u8 is_l3 = 0;
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600736 u32 instance = ~0;
Ed Warnickecb9cada2015-12-08 15:45:58 -0700737 u32 encap_fib_index = 0;
Eyal Baric5b13602016-11-24 19:42:43 +0200738 u32 mcast_sw_if_index = ~0;
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800739 u32 decap_next_index = VXLAN_INPUT_NEXT_L2_INPUT;
Ed Warnickecb9cada2015-12-08 15:45:58 -0700740 u32 vni = 0;
Artem Glazychev839dcc02020-12-01 02:39:21 +0700741 u32 src_port = 0;
742 u32 dst_port = 0;
Eyal Bariaa0180b2018-03-27 11:43:57 +0300743 u32 table_id;
Eyal Bari0fa56782018-06-04 12:25:05 +0300744 clib_error_t *parse_error = NULL;
Eyal Baric5b13602016-11-24 19:42:43 +0200745
Ed Warnickecb9cada2015-12-08 15:45:58 -0700746 /* Get a line of input. */
Eyal Bari8f576482018-05-07 10:13:44 +0300747 if (!unformat_user (input, unformat_line_input, line_input))
Ed Warnickecb9cada2015-12-08 15:45:58 -0700748 return 0;
749
Eyal Bari8f576482018-05-07 10:13:44 +0300750 while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
751 {
752 if (unformat (line_input, "del"))
753 {
754 is_add = 0;
755 }
756 else if (unformat (line_input, "instance %d", &instance))
757 ;
758 else if (unformat (line_input, "src %U",
759 unformat_ip46_address, &src, IP46_TYPE_ANY))
760 {
761 src_set = 1;
762 ip46_address_is_ip4 (&src) ? (ipv4_set = 1) : (ipv6_set = 1);
763 }
764 else if (unformat (line_input, "dst %U",
765 unformat_ip46_address, &dst, IP46_TYPE_ANY))
766 {
767 dst_set = 1;
768 ip46_address_is_ip4 (&dst) ? (ipv4_set = 1) : (ipv6_set = 1);
769 }
770 else if (unformat (line_input, "group %U %U",
771 unformat_ip46_address, &dst, IP46_TYPE_ANY,
772 unformat_vnet_sw_interface,
773 vnet_get_main (), &mcast_sw_if_index))
774 {
775 grp_set = dst_set = 1;
776 ip46_address_is_ip4 (&dst) ? (ipv4_set = 1) : (ipv6_set = 1);
777 }
778 else if (unformat (line_input, "encap-vrf-id %d", &table_id))
779 {
780 encap_fib_index =
781 fib_table_find (fib_ip_proto (ipv6_set), table_id);
Eyal Bari8f576482018-05-07 10:13:44 +0300782 }
Nathan Skrzypczak3e384222021-04-21 19:56:34 +0200783 else if (unformat (line_input, "l3"))
784 is_l3 = 1;
Eyal Bari8f576482018-05-07 10:13:44 +0300785 else if (unformat (line_input, "decap-next %U", unformat_decap_next,
786 &decap_next_index, ipv4_set))
787 ;
788 else if (unformat (line_input, "vni %d", &vni))
789 ;
Artem Glazychev839dcc02020-12-01 02:39:21 +0700790 else if (unformat (line_input, "src_port %d", &src_port))
791 ;
792 else if (unformat (line_input, "dst_port %d", &dst_port))
793 ;
Eyal Bari8f576482018-05-07 10:13:44 +0300794 else
795 {
Eyal Bari0fa56782018-06-04 12:25:05 +0300796 parse_error = clib_error_return (0, "parse error: '%U'",
797 format_unformat_error, line_input);
Eyal Bari8f576482018-05-07 10:13:44 +0300798 break;
799 }
800 }
Ed Warnickecb9cada2015-12-08 15:45:58 -0700801
Eyal Bariaa0180b2018-03-27 11:43:57 +0300802 unformat_free (line_input);
803
Eyal Bari0fa56782018-06-04 12:25:05 +0300804 if (parse_error)
805 return parse_error;
806
Nathan Skrzypczak3e384222021-04-21 19:56:34 +0200807 if (is_l3 && decap_next_index == VXLAN_INPUT_NEXT_L2_INPUT)
808 {
809 vlib_node_t *node = vlib_get_node_by_name (
810 vm, (u8 *) (ipv4_set ? "ip4-input" : "ip6-input"));
811 decap_next_index = get_decap_next_for_node (node->index, ipv4_set);
812 }
813
Eyal Bari0fa56782018-06-04 12:25:05 +0300814 if (encap_fib_index == ~0)
815 return clib_error_return (0, "nonexistent encap-vrf-id %d", table_id);
Eyal Bariaa0180b2018-03-27 11:43:57 +0300816
Ed Warnickecb9cada2015-12-08 15:45:58 -0700817 if (src_set == 0)
Eyal Bariaa0180b2018-03-27 11:43:57 +0300818 return clib_error_return (0, "tunnel src address not specified");
Ed Warnickecb9cada2015-12-08 15:45:58 -0700819
820 if (dst_set == 0)
Eyal Bariaa0180b2018-03-27 11:43:57 +0300821 return clib_error_return (0, "tunnel dst address not specified");
Ed Warnickecb9cada2015-12-08 15:45:58 -0700822
Eyal Bari8f576482018-05-07 10:13:44 +0300823 if (grp_set && !ip46_address_is_multicast (&dst))
Eyal Bariaa0180b2018-03-27 11:43:57 +0300824 return clib_error_return (0, "tunnel group address not multicast");
Eyal Baric5b13602016-11-24 19:42:43 +0200825
Eyal Bari8f576482018-05-07 10:13:44 +0300826 if (grp_set == 0 && ip46_address_is_multicast (&dst))
Eyal Bariaa0180b2018-03-27 11:43:57 +0300827 return clib_error_return (0, "dst address must be unicast");
John Lo56912c82016-12-08 16:10:02 -0500828
Eyal Baric5b13602016-11-24 19:42:43 +0200829 if (grp_set && mcast_sw_if_index == ~0)
Eyal Bariaa0180b2018-03-27 11:43:57 +0300830 return clib_error_return (0, "tunnel nonexistent multicast device");
Eyal Baric5b13602016-11-24 19:42:43 +0200831
Chris Luke99cb3352016-04-26 10:49:53 -0400832 if (ipv4_set && ipv6_set)
Eyal Bariaa0180b2018-03-27 11:43:57 +0300833 return clib_error_return (0, "both IPv4 and IPv6 addresses specified");
Chris Luke99cb3352016-04-26 10:49:53 -0400834
Eyal Bari8f576482018-05-07 10:13:44 +0300835 if (ip46_address_cmp (&src, &dst) == 0)
Eyal Bariaa0180b2018-03-27 11:43:57 +0300836 return clib_error_return (0, "src and dst addresses are identical");
Chris Luke99cb3352016-04-26 10:49:53 -0400837
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800838 if (decap_next_index == ~0)
Eyal Bariaa0180b2018-03-27 11:43:57 +0300839 return clib_error_return (0, "next node not found");
Hongjun Nibeb4bf72016-11-25 00:03:46 +0800840
Ed Warnickecb9cada2015-12-08 15:45:58 -0700841 if (vni == 0)
Eyal Bariaa0180b2018-03-27 11:43:57 +0300842 return clib_error_return (0, "vni not specified");
Ed Warnickecb9cada2015-12-08 15:45:58 -0700843
Eyal Bari8f576482018-05-07 10:13:44 +0300844 if (vni >> 24)
Eyal Bariaa0180b2018-03-27 11:43:57 +0300845 return clib_error_return (0, "vni %d out of range", vni);
Ed Warnickecb9cada2015-12-08 15:45:58 -0700846
Nathan Skrzypczak3e384222021-04-21 19:56:34 +0200847 vnet_vxlan_add_del_tunnel_args_t a = { .is_add = is_add,
848 .is_ip6 = ipv6_set,
Nathan Skrzypczak3e384222021-04-21 19:56:34 +0200849 .instance = instance,
Eyal Bariaa0180b2018-03-27 11:43:57 +0300850#define _(x) .x = x,
Nathan Skrzypczak3e384222021-04-21 19:56:34 +0200851 foreach_copy_field
Ed Warnickecb9cada2015-12-08 15:45:58 -0700852#undef _
Eyal Bariaa0180b2018-03-27 11:43:57 +0300853 };
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600854
Eyal Bariaa0180b2018-03-27 11:43:57 +0300855 u32 tunnel_sw_if_index;
856 int rv = vnet_vxlan_add_del_tunnel (&a, &tunnel_sw_if_index);
Ed Warnickecb9cada2015-12-08 15:45:58 -0700857
Eyal Bari8f576482018-05-07 10:13:44 +0300858 switch (rv)
Ed Warnickecb9cada2015-12-08 15:45:58 -0700859 {
860 case 0:
Pierre Pfister78ea9c22016-05-23 12:51:54 +0100861 if (is_add)
Eyal Bari8f576482018-05-07 10:13:44 +0300862 vlib_cli_output (vm, "%U\n", format_vnet_sw_if_index_name,
863 vnet_get_main (), tunnel_sw_if_index);
Ed Warnickecb9cada2015-12-08 15:45:58 -0700864 break;
Ed Warnickecb9cada2015-12-08 15:45:58 -0700865
866 case VNET_API_ERROR_TUNNEL_EXIST:
Eyal Bariaa0180b2018-03-27 11:43:57 +0300867 return clib_error_return (0, "tunnel already exists...");
Ed Warnickecb9cada2015-12-08 15:45:58 -0700868
869 case VNET_API_ERROR_NO_SUCH_ENTRY:
Eyal Bariaa0180b2018-03-27 11:43:57 +0300870 return clib_error_return (0, "tunnel does not exist...");
Ed Warnickecb9cada2015-12-08 15:45:58 -0700871
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600872 case VNET_API_ERROR_INSTANCE_IN_USE:
Eyal Bariaa0180b2018-03-27 11:43:57 +0300873 return clib_error_return (0, "Instance is in use");
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600874
Ed Warnickecb9cada2015-12-08 15:45:58 -0700875 default:
Eyal Bariaa0180b2018-03-27 11:43:57 +0300876 return clib_error_return
Eyal Bari8f576482018-05-07 10:13:44 +0300877 (0, "vnet_vxlan_add_del_tunnel returned %d", rv);
Ed Warnickecb9cada2015-12-08 15:45:58 -0700878 }
879
Eyal Bari0fa56782018-06-04 12:25:05 +0300880 return 0;
Ed Warnickecb9cada2015-12-08 15:45:58 -0700881}
882
Billy McFallc5d9cda2016-09-14 10:39:58 -0400883/*?
884 * Add or delete a VXLAN Tunnel.
885 *
886 * VXLAN provides the features needed to allow L2 bridge domains (BDs)
887 * to span multiple servers. This is done by building an L2 overlay on
888 * top of an L3 network underlay using VXLAN tunnels.
889 *
890 * This makes it possible for servers to be co-located in the same data
891 * center or be separated geographically as long as they are reachable
892 * through the underlay L3 network.
893 *
894 * You can refer to this kind of L2 overlay bridge domain as a VXLAN
895 * (Virtual eXtensible VLAN) segment.
896 *
897 * @cliexpar
898 * Example of how to create a VXLAN Tunnel:
Artem Glazychev839dcc02020-12-01 02:39:21 +0700899 * @cliexcmd{create vxlan tunnel src 10.0.3.1 dst 10.0.3.3 vni 13 encap-vrf-id
900 7}
Jon Loeliger3d460bd2018-02-01 16:36:12 -0600901 * Example of how to create a VXLAN Tunnel with a known name, vxlan_tunnel42:
902 * @cliexcmd{create vxlan tunnel src 10.0.3.1 dst 10.0.3.3 instance 42}
Artem Glazychev839dcc02020-12-01 02:39:21 +0700903 * Example of how to create a multicast VXLAN Tunnel with a known name,
904 vxlan_tunnel23:
905 * @cliexcmd{create vxlan tunnel src 10.0.3.1 group 239.1.1.1
906 GigabitEthernet0/8/0 instance 23}
907 * Example of how to create a VXLAN Tunnel with custom udp-ports:
908 * @cliexcmd{create vxlan tunnel src 10.0.3.1 dst 10.0.3.3 vni 13 src_port
909 59000 dst_port 59001}
Billy McFallc5d9cda2016-09-14 10:39:58 -0400910 * Example of how to delete a VXLAN Tunnel:
911 * @cliexcmd{create vxlan tunnel src 10.0.3.1 dst 10.0.3.3 vni 13 del}
912 ?*/
913/* *INDENT-OFF* */
Ed Warnickecb9cada2015-12-08 15:45:58 -0700914VLIB_CLI_COMMAND (create_vxlan_tunnel_command, static) = {
915 .path = "create vxlan tunnel",
Eyal Bari8f576482018-05-07 10:13:44 +0300916 .short_help =
Artem Glazychev839dcc02020-12-01 02:39:21 +0700917 "create vxlan tunnel src <local-vtep-addr>"
918 " {dst <remote-vtep-addr>|group <mcast-vtep-addr> <intf-name>} vni <nn>"
919 " [instance <id>]"
Nathan Skrzypczak3e384222021-04-21 19:56:34 +0200920 " [encap-vrf-id <nn>] [decap-next [l2|node <name>]] [del] [l3]"
Artem Glazychev839dcc02020-12-01 02:39:21 +0700921 " [src_port <local-vtep-udp-port>] [dst_port <remote-vtep-udp-port>]",
Ed Warnickecb9cada2015-12-08 15:45:58 -0700922 .function = vxlan_add_del_tunnel_command_fn,
923};
Billy McFallc5d9cda2016-09-14 10:39:58 -0400924/* *INDENT-ON* */
Ed Warnickecb9cada2015-12-08 15:45:58 -0700925
926static clib_error_t *
927show_vxlan_tunnel_command_fn (vlib_main_t * vm,
Eyal Bari8f576482018-05-07 10:13:44 +0300928 unformat_input_t * input,
929 vlib_cli_command_t * cmd)
Ed Warnickecb9cada2015-12-08 15:45:58 -0700930{
Eyal Bari8f576482018-05-07 10:13:44 +0300931 vxlan_main_t *vxm = &vxlan_main;
932 vxlan_tunnel_t *t;
Eyal Bari0fa56782018-06-04 12:25:05 +0300933 int raw = 0;
Eyal Bari0fa56782018-06-04 12:25:05 +0300934
Neale Ranns16be62e2018-07-30 11:59:22 -0700935 while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
Eyal Bari0fa56782018-06-04 12:25:05 +0300936 {
Neale Ranns16be62e2018-07-30 11:59:22 -0700937 if (unformat (input, "raw"))
Eyal Bari0fa56782018-06-04 12:25:05 +0300938 raw = 1;
939 else
Neale Ranns16be62e2018-07-30 11:59:22 -0700940 return clib_error_return (0, "parse error: '%U'",
941 format_unformat_error, input);
Eyal Bari0fa56782018-06-04 12:25:05 +0300942 }
943
Ed Warnickecb9cada2015-12-08 15:45:58 -0700944 if (pool_elts (vxm->tunnels) == 0)
945 vlib_cli_output (vm, "No vxlan tunnels configured...");
946
Eyal Bari8f576482018-05-07 10:13:44 +0300947/* *INDENT-OFF* */
Damjan Marionb2c31b62020-12-13 21:47:40 +0100948 pool_foreach (t, vxm->tunnels)
949 {
Ed Warnickecb9cada2015-12-08 15:45:58 -0700950 vlib_cli_output (vm, "%U", format_vxlan_tunnel, t);
Damjan Marionb2c31b62020-12-13 21:47:40 +0100951 }
Eyal Bari8f576482018-05-07 10:13:44 +0300952/* *INDENT-ON* */
953
Eyal Bari0fa56782018-06-04 12:25:05 +0300954 if (raw)
Eyal Baridd47eca2018-07-08 08:15:56 +0300955 {
956 vlib_cli_output (vm, "Raw IPv4 Hash Table:\n%U\n",
957 format_bihash_16_8, &vxm->vxlan4_tunnel_by_key,
958 1 /* verbose */ );
959 vlib_cli_output (vm, "Raw IPv6 Hash Table:\n%U\n",
960 format_bihash_24_8, &vxm->vxlan6_tunnel_by_key,
961 1 /* verbose */ );
962 }
Eyal Bari0fa56782018-06-04 12:25:05 +0300963
Ed Warnickecb9cada2015-12-08 15:45:58 -0700964 return 0;
965}
966
Billy McFallc5d9cda2016-09-14 10:39:58 -0400967/*?
968 * Display all the VXLAN Tunnel entries.
969 *
970 * @cliexpar
971 * Example of how to display the VXLAN Tunnel entries:
972 * @cliexstart{show vxlan tunnel}
Artem Glazychev839dcc02020-12-01 02:39:21 +0700973 * [0] src 10.0.3.1 dst 10.0.3.3 src_port 4789 dst_port 4789 vni 13
974 encap_fib_index 0 sw_if_index 5 decap_next l2
Billy McFallc5d9cda2016-09-14 10:39:58 -0400975 * @cliexend
976 ?*/
977/* *INDENT-OFF* */
Ed Warnickecb9cada2015-12-08 15:45:58 -0700978VLIB_CLI_COMMAND (show_vxlan_tunnel_command, static) = {
979 .path = "show vxlan tunnel",
Eyal Bari0fa56782018-06-04 12:25:05 +0300980 .short_help = "show vxlan tunnel [raw]",
Ed Warnickecb9cada2015-12-08 15:45:58 -0700981 .function = show_vxlan_tunnel_command_fn,
982};
Billy McFallc5d9cda2016-09-14 10:39:58 -0400983/* *INDENT-ON* */
Ed Warnickecb9cada2015-12-08 15:45:58 -0700984
Chris Luke99cb3352016-04-26 10:49:53 -0400985
Eyal Bari8f576482018-05-07 10:13:44 +0300986void
987vnet_int_vxlan_bypass_mode (u32 sw_if_index, u8 is_ip6, u8 is_enable)
John Lo2b81eb82017-01-30 13:12:10 -0500988{
Jon Loeligere3034cd2019-01-03 12:56:02 -0600989 vxlan_main_t *vxm = &vxlan_main;
990
991 if (pool_is_free_index (vxm->vnet_main->interface_main.sw_interfaces,
992 sw_if_index))
993 return;
994
995 is_enable = ! !is_enable;
996
John Lo2b81eb82017-01-30 13:12:10 -0500997 if (is_ip6)
Jon Loeligere3034cd2019-01-03 12:56:02 -0600998 {
999 if (clib_bitmap_get (vxm->bm_ip6_bypass_enabled_by_sw_if, sw_if_index)
1000 != is_enable)
1001 {
1002 vnet_feature_enable_disable ("ip6-unicast", "ip6-vxlan-bypass",
1003 sw_if_index, is_enable, 0, 0);
1004 vxm->bm_ip6_bypass_enabled_by_sw_if =
1005 clib_bitmap_set (vxm->bm_ip6_bypass_enabled_by_sw_if,
1006 sw_if_index, is_enable);
1007 }
1008 }
John Lo2b81eb82017-01-30 13:12:10 -05001009 else
Jon Loeligere3034cd2019-01-03 12:56:02 -06001010 {
1011 if (clib_bitmap_get (vxm->bm_ip4_bypass_enabled_by_sw_if, sw_if_index)
1012 != is_enable)
1013 {
1014 vnet_feature_enable_disable ("ip4-unicast", "ip4-vxlan-bypass",
1015 sw_if_index, is_enable, 0, 0);
1016 vxm->bm_ip4_bypass_enabled_by_sw_if =
1017 clib_bitmap_set (vxm->bm_ip4_bypass_enabled_by_sw_if,
1018 sw_if_index, is_enable);
1019 }
1020 }
John Lo2b81eb82017-01-30 13:12:10 -05001021}
1022
1023
1024static clib_error_t *
1025set_ip_vxlan_bypass (u32 is_ip6,
Eyal Bari8f576482018-05-07 10:13:44 +03001026 unformat_input_t * input, vlib_cli_command_t * cmd)
John Lo2b81eb82017-01-30 13:12:10 -05001027{
Eyal Bari8f576482018-05-07 10:13:44 +03001028 unformat_input_t _line_input, *line_input = &_line_input;
1029 vnet_main_t *vnm = vnet_get_main ();
1030 clib_error_t *error = 0;
John Lo2b81eb82017-01-30 13:12:10 -05001031 u32 sw_if_index, is_enable;
1032
1033 sw_if_index = ~0;
1034 is_enable = 1;
1035
Eyal Bari8f576482018-05-07 10:13:44 +03001036 if (!unformat_user (input, unformat_line_input, line_input))
John Lo2b81eb82017-01-30 13:12:10 -05001037 return 0;
1038
1039 while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
1040 {
Eyal Bari8f576482018-05-07 10:13:44 +03001041 if (unformat_user
1042 (line_input, unformat_vnet_sw_interface, vnm, &sw_if_index))
1043 ;
John Lo2b81eb82017-01-30 13:12:10 -05001044 else if (unformat (line_input, "del"))
Eyal Bari8f576482018-05-07 10:13:44 +03001045 is_enable = 0;
John Lo2b81eb82017-01-30 13:12:10 -05001046 else
Eyal Bari8f576482018-05-07 10:13:44 +03001047 {
John Lo2b81eb82017-01-30 13:12:10 -05001048 error = unformat_parse_error (line_input);
1049 goto done;
1050 }
1051 }
1052
1053 if (~0 == sw_if_index)
1054 {
1055 error = clib_error_return (0, "unknown interface `%U'",
1056 format_unformat_error, line_input);
1057 goto done;
1058 }
1059
1060 vnet_int_vxlan_bypass_mode (sw_if_index, is_ip6, is_enable);
1061
Eyal Bari8f576482018-05-07 10:13:44 +03001062done:
Billy McFalla9a20e72017-02-15 11:39:12 -05001063 unformat_free (line_input);
1064
John Lo2b81eb82017-01-30 13:12:10 -05001065 return error;
1066}
1067
1068static clib_error_t *
1069set_ip4_vxlan_bypass (vlib_main_t * vm,
Eyal Bari8f576482018-05-07 10:13:44 +03001070 unformat_input_t * input, vlib_cli_command_t * cmd)
John Lo2b81eb82017-01-30 13:12:10 -05001071{
1072 return set_ip_vxlan_bypass (0, input, cmd);
1073}
1074
1075/*?
Eyal Bari8f576482018-05-07 10:13:44 +03001076 * This command adds the 'ip4-vxlan-bypass' graph node for a given interface.
John Lo2b81eb82017-01-30 13:12:10 -05001077 * By adding the IPv4 vxlan-bypass graph node to an interface, the node checks
Nathan Skrzypczak5e6a1652021-10-06 15:03:35 +02001078 * for and validate input vxlan packet and bypass ip4-lookup, ip4-local,
Eyal Bari8f576482018-05-07 10:13:44 +03001079 * ip4-udp-lookup nodes to speedup vxlan packet forwarding. This node will
John Lo2b81eb82017-01-30 13:12:10 -05001080 * cause extra overhead to for non-vxlan packets which is kept at a minimum.
1081 *
1082 * @cliexpar
1083 * @parblock
1084 * Example of graph node before ip4-vxlan-bypass is enabled:
1085 * @cliexstart{show vlib graph ip4-vxlan-bypass}
1086 * Name Next Previous
1087 * ip4-vxlan-bypass error-drop [0]
1088 * vxlan4-input [1]
Eyal Bari8f576482018-05-07 10:13:44 +03001089 * ip4-lookup [2]
John Lo2b81eb82017-01-30 13:12:10 -05001090 * @cliexend
1091 *
1092 * Example of how to enable ip4-vxlan-bypass on an interface:
1093 * @cliexcmd{set interface ip vxlan-bypass GigabitEthernet2/0/0}
1094 *
1095 * Example of graph node after ip4-vxlan-bypass is enabled:
1096 * @cliexstart{show vlib graph ip4-vxlan-bypass}
Eyal Bari8f576482018-05-07 10:13:44 +03001097 * Name Next Previous
1098 * ip4-vxlan-bypass error-drop [0] ip4-input
1099 * vxlan4-input [1] ip4-input-no-checksum
1100 * ip4-lookup [2]
John Lo2b81eb82017-01-30 13:12:10 -05001101 * @cliexend
1102 *
Paul Vinciguerrabdc0e6b2018-09-22 05:32:50 -07001103 * Example of how to display the feature enabled on an interface:
John Lo2b81eb82017-01-30 13:12:10 -05001104 * @cliexstart{show ip interface features GigabitEthernet2/0/0}
1105 * IP feature paths configured on GigabitEthernet2/0/0...
1106 * ...
1107 * ipv4 unicast:
1108 * ip4-vxlan-bypass
1109 * ip4-lookup
1110 * ...
1111 * @cliexend
1112 *
1113 * Example of how to disable ip4-vxlan-bypass on an interface:
1114 * @cliexcmd{set interface ip vxlan-bypass GigabitEthernet2/0/0 del}
1115 * @endparblock
1116?*/
1117/* *INDENT-OFF* */
1118VLIB_CLI_COMMAND (set_interface_ip_vxlan_bypass_command, static) = {
1119 .path = "set interface ip vxlan-bypass",
1120 .function = set_ip4_vxlan_bypass,
1121 .short_help = "set interface ip vxlan-bypass <interface> [del]",
1122};
1123/* *INDENT-ON* */
1124
1125static clib_error_t *
1126set_ip6_vxlan_bypass (vlib_main_t * vm,
Eyal Bari8f576482018-05-07 10:13:44 +03001127 unformat_input_t * input, vlib_cli_command_t * cmd)
John Lo2b81eb82017-01-30 13:12:10 -05001128{
1129 return set_ip_vxlan_bypass (1, input, cmd);
1130}
1131
1132/*?
Eyal Bari8f576482018-05-07 10:13:44 +03001133 * This command adds the 'ip6-vxlan-bypass' graph node for a given interface.
John Lo2b81eb82017-01-30 13:12:10 -05001134 * By adding the IPv6 vxlan-bypass graph node to an interface, the node checks
Nathan Skrzypczak5e6a1652021-10-06 15:03:35 +02001135 * for and validate input vxlan packet and bypass ip6-lookup, ip6-local,
Eyal Bari8f576482018-05-07 10:13:44 +03001136 * ip6-udp-lookup nodes to speedup vxlan packet forwarding. This node will
John Lo2b81eb82017-01-30 13:12:10 -05001137 * cause extra overhead to for non-vxlan packets which is kept at a minimum.
1138 *
1139 * @cliexpar
1140 * @parblock
1141 * Example of graph node before ip6-vxlan-bypass is enabled:
1142 * @cliexstart{show vlib graph ip6-vxlan-bypass}
1143 * Name Next Previous
1144 * ip6-vxlan-bypass error-drop [0]
1145 * vxlan6-input [1]
Eyal Bari8f576482018-05-07 10:13:44 +03001146 * ip6-lookup [2]
John Lo2b81eb82017-01-30 13:12:10 -05001147 * @cliexend
1148 *
1149 * Example of how to enable ip6-vxlan-bypass on an interface:
1150 * @cliexcmd{set interface ip6 vxlan-bypass GigabitEthernet2/0/0}
1151 *
1152 * Example of graph node after ip6-vxlan-bypass is enabled:
1153 * @cliexstart{show vlib graph ip6-vxlan-bypass}
Eyal Bari8f576482018-05-07 10:13:44 +03001154 * Name Next Previous
1155 * ip6-vxlan-bypass error-drop [0] ip6-input
1156 * vxlan6-input [1] ip4-input-no-checksum
1157 * ip6-lookup [2]
John Lo2b81eb82017-01-30 13:12:10 -05001158 * @cliexend
1159 *
Paul Vinciguerrabdc0e6b2018-09-22 05:32:50 -07001160 * Example of how to display the feature enabled on an interface:
John Lo2b81eb82017-01-30 13:12:10 -05001161 * @cliexstart{show ip interface features GigabitEthernet2/0/0}
1162 * IP feature paths configured on GigabitEthernet2/0/0...
1163 * ...
1164 * ipv6 unicast:
1165 * ip6-vxlan-bypass
1166 * ip6-lookup
1167 * ...
1168 * @cliexend
1169 *
1170 * Example of how to disable ip6-vxlan-bypass on an interface:
1171 * @cliexcmd{set interface ip6 vxlan-bypass GigabitEthernet2/0/0 del}
1172 * @endparblock
1173?*/
1174/* *INDENT-OFF* */
1175VLIB_CLI_COMMAND (set_interface_ip6_vxlan_bypass_command, static) = {
1176 .path = "set interface ip6 vxlan-bypass",
1177 .function = set_ip6_vxlan_bypass,
Paul Vinciguerra5fb22782019-11-07 17:20:48 -05001178 .short_help = "set interface ip6 vxlan-bypass <interface> [del]",
John Lo2b81eb82017-01-30 13:12:10 -05001179};
1180/* *INDENT-ON* */
1181
eyal bariaf86a482018-04-17 11:20:27 +03001182int
1183vnet_vxlan_add_del_rx_flow (u32 hw_if_index, u32 t_index, int is_add)
1184{
1185 vxlan_main_t *vxm = &vxlan_main;
1186 vxlan_tunnel_t *t = pool_elt_at_index (vxm->tunnels, t_index);
1187 vnet_main_t *vnm = vnet_get_main ();
1188 if (is_add)
1189 {
1190 if (t->flow_index == ~0)
1191 {
1192 vxlan_main_t *vxm = &vxlan_main;
1193 vnet_flow_t flow = {
1194 .actions =
Chenmin Sun1ec9fdb2019-12-05 01:41:35 +08001195 VNET_FLOW_ACTION_REDIRECT_TO_NODE | VNET_FLOW_ACTION_MARK |
1196 VNET_FLOW_ACTION_BUFFER_ADVANCE,
eyal bariaf86a482018-04-17 11:20:27 +03001197 .mark_flow_id = t->dev_instance + vxm->flow_id_start,
1198 .redirect_node_index = vxlan4_flow_input_node.index,
Chenmin Sun1ec9fdb2019-12-05 01:41:35 +08001199 .buffer_advance = sizeof (ethernet_header_t),
eyal bariaf86a482018-04-17 11:20:27 +03001200 .type = VNET_FLOW_TYPE_IP4_VXLAN,
1201 .ip4_vxlan = {
Chenmin Sun34bfa502020-07-27 17:40:17 +08001202 .protocol.prot = IP_PROTOCOL_UDP,
1203 .src_addr.addr = t->dst.ip4,
1204 .dst_addr.addr = t->src.ip4,
1205 .src_addr.mask.as_u32 = ~0,
1206 .dst_addr.mask.as_u32 = ~0,
Artem Glazychev839dcc02020-12-01 02:39:21 +07001207 .dst_port.port = t->src_port,
Chenmin Sun34bfa502020-07-27 17:40:17 +08001208 .dst_port.mask = 0xFF,
eyal bariaf86a482018-04-17 11:20:27 +03001209 .vni = t->vni,
1210 }
1211 ,
1212 };
1213 vnet_flow_add (vnm, &flow, &t->flow_index);
1214 }
1215 return vnet_flow_enable (vnm, t->flow_index, hw_if_index);
1216 }
1217 /* flow index is removed when the tunnel is deleted */
1218 return vnet_flow_disable (vnm, t->flow_index, hw_if_index);
1219}
1220
1221u32
1222vnet_vxlan_get_tunnel_index (u32 sw_if_index)
1223{
1224 vxlan_main_t *vxm = &vxlan_main;
1225
Eyal Baricd307742018-07-22 12:45:15 +03001226 if (sw_if_index >= vec_len (vxm->tunnel_index_by_sw_if_index))
eyal bariaf86a482018-04-17 11:20:27 +03001227 return ~0;
1228 return vxm->tunnel_index_by_sw_if_index[sw_if_index];
1229}
1230
1231static clib_error_t *
1232vxlan_offload_command_fn (vlib_main_t * vm,
1233 unformat_input_t * input, vlib_cli_command_t * cmd)
1234{
1235 unformat_input_t _line_input, *line_input = &_line_input;
1236
1237 /* Get a line of input. */
1238 if (!unformat_user (input, unformat_line_input, line_input))
1239 return 0;
1240
1241 vnet_main_t *vnm = vnet_get_main ();
1242 u32 rx_sw_if_index = ~0;
1243 u32 hw_if_index = ~0;
1244 int is_add = 1;
1245
1246 while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
1247 {
1248 if (unformat (line_input, "hw %U", unformat_vnet_hw_interface, vnm,
1249 &hw_if_index))
1250 continue;
1251 if (unformat (line_input, "rx %U", unformat_vnet_sw_interface, vnm,
1252 &rx_sw_if_index))
1253 continue;
1254 if (unformat (line_input, "del"))
1255 {
1256 is_add = 0;
1257 continue;
1258 }
1259 return clib_error_return (0, "unknown input `%U'",
1260 format_unformat_error, line_input);
1261 }
1262
1263 if (rx_sw_if_index == ~0)
1264 return clib_error_return (0, "missing rx interface");
1265 if (hw_if_index == ~0)
1266 return clib_error_return (0, "missing hw interface");
1267
1268 u32 t_index = vnet_vxlan_get_tunnel_index (rx_sw_if_index);;
1269 if (t_index == ~0)
1270 return clib_error_return (0, "%U is not a vxlan tunnel",
1271 format_vnet_sw_if_index_name, vnm,
1272 rx_sw_if_index);
1273
1274 vxlan_main_t *vxm = &vxlan_main;
1275 vxlan_tunnel_t *t = pool_elt_at_index (vxm->tunnels, t_index);
1276
1277 if (!ip46_address_is_ip4 (&t->dst))
1278 return clib_error_return (0, "currently only IPV4 tunnels are supported");
1279
1280 vnet_hw_interface_t *hw_if = vnet_get_hw_interface (vnm, hw_if_index);
1281 ip4_main_t *im = &ip4_main;
1282 u32 rx_fib_index =
1283 vec_elt (im->fib_index_by_sw_if_index, hw_if->sw_if_index);
1284
1285 if (t->encap_fib_index != rx_fib_index)
1286 return clib_error_return (0, "interface/tunnel fib mismatch");
1287
1288 if (vnet_vxlan_add_del_rx_flow (hw_if_index, t_index, is_add))
1289 return clib_error_return (0, "error %s flow",
1290 is_add ? "enabling" : "disabling");
1291
1292 return 0;
1293}
1294
1295/* *INDENT-OFF* */
1296VLIB_CLI_COMMAND (vxlan_offload_command, static) = {
1297 .path = "set flow-offload vxlan",
1298 .short_help =
Paul Vinciguerrabdc0e6b2018-09-22 05:32:50 -07001299 "set flow-offload vxlan hw <interface-name> rx <tunnel-name> [del]",
eyal bariaf86a482018-04-17 11:20:27 +03001300 .function = vxlan_offload_command_fn,
1301};
1302/* *INDENT-ON* */
1303
Eyal Bari0fa56782018-06-04 12:25:05 +03001304#define VXLAN_HASH_NUM_BUCKETS (2 * 1024)
1305#define VXLAN_HASH_MEMORY_SIZE (1 << 20)
1306
Eyal Bari8f576482018-05-07 10:13:44 +03001307clib_error_t *
1308vxlan_init (vlib_main_t * vm)
Ed Warnickecb9cada2015-12-08 15:45:58 -07001309{
Eyal Bari8f576482018-05-07 10:13:44 +03001310 vxlan_main_t *vxm = &vxlan_main;
1311
1312 vxm->vnet_main = vnet_get_main ();
Ed Warnickecb9cada2015-12-08 15:45:58 -07001313 vxm->vlib_main = vm;
1314
eyal bariaf86a482018-04-17 11:20:27 +03001315 vnet_flow_get_range (vxm->vnet_main, "vxlan", 1024 * 1024,
1316 &vxm->flow_id_start);
1317
Jon Loeligere3034cd2019-01-03 12:56:02 -06001318 vxm->bm_ip4_bypass_enabled_by_sw_if = 0;
1319 vxm->bm_ip6_bypass_enabled_by_sw_if = 0;
1320
Chris Luke99cb3352016-04-26 10:49:53 -04001321 /* initialize the ip6 hash */
Eyal Baridd47eca2018-07-08 08:15:56 +03001322 clib_bihash_init_16_8 (&vxm->vxlan4_tunnel_by_key, "vxlan4",
1323 VXLAN_HASH_NUM_BUCKETS, VXLAN_HASH_MEMORY_SIZE);
1324 clib_bihash_init_24_8 (&vxm->vxlan6_tunnel_by_key, "vxlan6",
Eyal Bari0fa56782018-06-04 12:25:05 +03001325 VXLAN_HASH_NUM_BUCKETS, VXLAN_HASH_MEMORY_SIZE);
Nick Zavaritsky27518c22020-02-27 15:54:58 +00001326 vxm->vtep_table = vtep_table_create ();
Eyal Bari8f576482018-05-07 10:13:44 +03001327 vxm->mcast_shared = hash_create_mem (0,
1328 sizeof (ip46_address_t),
1329 sizeof (mcast_shared_t));
Chris Luke99cb3352016-04-26 10:49:53 -04001330
Eyal Bari8f576482018-05-07 10:13:44 +03001331 fib_node_register_type (FIB_NODE_TYPE_VXLAN_TUNNEL, &vxlan_vft);
John Loc42912d2016-11-07 18:30:47 -05001332
Ed Warnickecb9cada2015-12-08 15:45:58 -07001333 return 0;
1334}
1335
Eyal Bari8f576482018-05-07 10:13:44 +03001336VLIB_INIT_FUNCTION (vxlan_init);
Jon Loeliger3d460bd2018-02-01 16:36:12 -06001337
1338/*
Eyal Bari8f576482018-05-07 10:13:44 +03001339 * fd.io coding-style-patch-verification: ON
1340 *
Jon Loeliger3d460bd2018-02-01 16:36:12 -06001341 * Local Variables:
1342 * eval: (c-set-style "gnu")
1343 * End:
1344 */