summaryrefslogtreecommitdiffstats
path: root/src/plugins/lb/lb.c
diff options
context:
space:
mode:
authorHongjun Ni <hongjun.ni@intel.com>2018-01-23 19:17:23 +0800
committerHongjun Ni <hongjun.ni@intel.com>2018-02-02 02:20:46 +0000
commit647f609a11e2afb91a5216ca99d0705a3e1212a7 (patch)
tree149ab0443d42a8d8a7620c8bce917ec87edf1ae7 /src/plugins/lb/lb.c
parent91389ac2c28ae10f2b7f766e4dfe7a7fd96dc5e0 (diff)
Add L3DSR feature in LB plugin
L3DSR is used to overcome Layer 2 limitations of Direct Server Return Load Balancing. It maps VIP to DSCP bits, and reuse TOS bits to transfer it to server, and then server will get VIP from DSCP-to-VIP mapping. Please refer to https://www.nanog.org/meetings/nanog51/presentations/Monday/NANOG51.Talk45.nanog51-Schaumann.pdf Change-Id: I403ffeadfb04ed0265086eb2dc41f2e17f8f34cb Signed-off-by: Hongjun Ni <hongjun.ni@intel.com>
Diffstat (limited to 'src/plugins/lb/lb.c')
-rw-r--r--src/plugins/lb/lb.c60
1 files changed, 53 insertions, 7 deletions
diff --git a/src/plugins/lb/lb.c b/src/plugins/lb/lb.c
index fee88056eb4..06953a45aaa 100644
--- a/src/plugins/lb/lb.c
+++ b/src/plugins/lb/lb.c
@@ -48,6 +48,12 @@ const static char* const * const lb_dpo_gre6_nodes[DPO_PROTO_NUM] =
[DPO_PROTO_IP6] = lb_dpo_gre6_ip6,
};
+const static char * const lb_dpo_l3dsr_ip4[] = { "lb4-l3dsr" , NULL };
+const static char* const * const lb_dpo_l3dsr_nodes[DPO_PROTO_NUM] =
+ {
+ [DPO_PROTO_IP4] = lb_dpo_l3dsr_ip4,
+ };
+
u32 lb_hash_time_now(vlib_main_t * vm)
{
return (u32) (vlib_time_now(vm) + 10000);
@@ -81,6 +87,7 @@ static char *lb_vip_type_strings[] = {
[LB_VIP_TYPE_IP6_GRE4] = "ip6-gre4",
[LB_VIP_TYPE_IP4_GRE6] = "ip4-gre6",
[LB_VIP_TYPE_IP4_GRE4] = "ip4-gre4",
+ [LB_VIP_TYPE_IP4_L3DSR] = "ip4-l3dsr",
};
u8 *format_lb_vip_type (u8 * s, va_list * args)
@@ -140,6 +147,13 @@ u8 *format_lb_vip_detailed (u8 * s, va_list * args)
format_white_space, indent,
vip->new_flow_table_mask + 1);
+ if (vip->type == LB_VIP_TYPE_IP4_L3DSR)
+ {
+ s = format(s, "%U dscp:%u\n",
+ format_white_space, indent,
+ vip->dscp);
+ }
+
//Print counters
s = format(s, "%U counters:\n",
format_white_space, indent);
@@ -434,7 +448,7 @@ int lb_vip_add_ass(u32 vip_index, ip46_address_t *addresses, u32 n)
return VNET_API_ERROR_NO_SUCH_ENTRY;
}
- ip46_type_t type = lb_vip_is_gre4(vip)?IP46_TYPE_IP4:IP46_TYPE_IP6;
+ ip46_type_t type = lb_encap_is_ip4(vip)?IP46_TYPE_IP4:IP46_TYPE_IP6;
u32 *to_be_added = 0;
u32 *to_be_updated = 0;
u32 i;
@@ -497,7 +511,7 @@ next:
* so we are informed when its forwarding changes
*/
fib_prefix_t nh = {};
- if (lb_vip_is_gre4(vip)) {
+ if (lb_encap_is_ip4(vip)) {
nh.fp_addr.ip4 = as->address.ip4;
nh.fp_len = 32;
nh.fp_proto = FIB_PROTOCOL_IP4;
@@ -595,6 +609,8 @@ int lb_vip_del_ass(u32 vip_index, ip46_address_t *addresses, u32 n)
static void lb_vip_add_adjacency(lb_main_t *lbm, lb_vip_t *vip)
{
dpo_proto_t proto = 0;
+ dpo_type_t dpo_type = 0;
+
dpo_id_t dpo = DPO_INVALID;
fib_prefix_t pfx = {};
if (lb_vip_is_ip4(vip)) {
@@ -608,8 +624,15 @@ static void lb_vip_add_adjacency(lb_main_t *lbm, lb_vip_t *vip)
pfx.fp_proto = FIB_PROTOCOL_IP6;
proto = DPO_PROTO_IP6;
}
- dpo_set(&dpo, lb_vip_is_gre4(vip)?lbm->dpo_gre4_type:lbm->dpo_gre6_type,
- proto, vip - lbm->vips);
+
+ if(lb_vip_is_gre4(vip))
+ dpo_type = lbm->dpo_gre4_type;
+ else if (lb_vip_is_gre6(vip))
+ dpo_type = lbm->dpo_gre6_type;
+ else if (lb_vip_is_l3dsr(vip))
+ dpo_type = lbm->dpo_l3dsr_type;
+
+ dpo_set(&dpo, dpo_type, proto, vip - lbm->vips);
fib_table_entry_special_dpo_add(0,
&pfx,
FIB_SOURCE_PLUGIN_HI,
@@ -636,10 +659,12 @@ static void lb_vip_del_adjacency(lb_main_t *lbm, lb_vip_t *vip)
fib_table_entry_special_remove(0, &pfx, FIB_SOURCE_PLUGIN_HI);
}
-int lb_vip_add(ip46_address_t *prefix, u8 plen, lb_vip_type_t type, u32 new_length, u32 *vip_index)
+int lb_vip_add(ip46_address_t *prefix, u8 plen, lb_vip_type_t type, u8 dscp,
+ u32 new_length, u32 *vip_index)
{
lb_main_t *lbm = &lb_main;
lb_vip_t *vip;
+
lb_get_writer_lock();
ip46_prefix_normalize(prefix, plen);
@@ -655,9 +680,19 @@ int lb_vip_add(ip46_address_t *prefix, u8 plen, lb_vip_type_t type, u32 new_leng
if (ip46_prefix_is_ip4(prefix, plen) &&
(type != LB_VIP_TYPE_IP4_GRE4) &&
- (type != LB_VIP_TYPE_IP4_GRE6))
+ (type != LB_VIP_TYPE_IP4_GRE6) &&
+ (type != LB_VIP_TYPE_IP4_L3DSR))
+ return VNET_API_ERROR_INVALID_ADDRESS_FAMILY;
+
+ if ((!ip46_prefix_is_ip4(prefix, plen)) &&
+ (type != LB_VIP_TYPE_IP6_GRE4) &&
+ (type != LB_VIP_TYPE_IP6_GRE6))
return VNET_API_ERROR_INVALID_ADDRESS_FAMILY;
+ if ((type == LB_VIP_TYPE_IP4_L3DSR) && (dscp >= 64 ) )
+ {
+ return VNET_API_ERROR_VALUE_EXIST;
+ }
//Allocate
pool_get(lbm->vips, vip);
@@ -667,6 +702,7 @@ int lb_vip_add(ip46_address_t *prefix, u8 plen, lb_vip_type_t type, u32 new_leng
vip->plen = plen;
vip->last_garbage_collection = (u32) vlib_time_now(vlib_get_main());
vip->type = type;
+ vip->dscp = dscp;
vip->flags = LB_VIP_FLAGS_USED;
vip->as_indexes = 0;
@@ -775,7 +811,16 @@ lb_as_stack (lb_as_t *as)
{
lb_main_t *lbm = &lb_main;
lb_vip_t *vip = &lbm->vips[as->vip_index];
- dpo_stack(lb_vip_is_gre4(vip)?lbm->dpo_gre4_type:lbm->dpo_gre6_type,
+ dpo_type_t dpo_type = 0;
+
+ if(lb_vip_is_gre4(vip))
+ dpo_type = lbm->dpo_gre4_type;
+ else if (lb_vip_is_gre6(vip))
+ dpo_type = lbm->dpo_gre6_type;
+ else if (lb_vip_is_l3dsr(vip))
+ dpo_type = lbm->dpo_l3dsr_type;
+
+ dpo_stack(dpo_type,
lb_vip_is_ip4(vip)?DPO_PROTO_IP4:DPO_PROTO_IP6,
&as->dpo,
fib_entry_contribute_ip_forwarding(
@@ -819,6 +864,7 @@ lb_init (vlib_main_t * vm)
lbm->ip6_src_address.as_u64[1] = 0xffffffffffffffffL;
lbm->dpo_gre4_type = dpo_register_new_type(&lb_vft, lb_dpo_gre4_nodes);
lbm->dpo_gre6_type = dpo_register_new_type(&lb_vft, lb_dpo_gre6_nodes);
+ lbm->dpo_l3dsr_type = dpo_register_new_type(&lb_vft, lb_dpo_l3dsr_nodes);
lbm->fib_node_type = fib_node_register_new_type(&lb_fib_node_vft);
//Init AS reference counters