/**
* adj_mcast_update_rewrite
*
* Update the adjacency's rewrite string. A NULL string implies the
* rewrite is reset (i.e. when ARP/ND entry is gone).
* NB: the adj being updated may be handling traffic in the DP.
*/
void
adj_mcast_update_rewrite (adj_index_t adj_index,
u8 *rewrite,
u8 offset)
{
ip_adjacency_t *adj;
ASSERT(ADJ_INDEX_INVALID != adj_index);
adj = adj_get(adj_index);
/*
* update the adj's rewrite string and build the arc
* from the rewrite node to the interface's TX node
*/
adj_nbr_update_rewrite_internal(adj, IP_LOOKUP_NEXT_MCAST,
adj_get_mcast_node(adj->ia_nh_proto),
vnet_tx_node_index_for_sw_interface(
vnet_get_main(),
adj->rewrite_header.sw_if_index),
rewrite);
/*
* set the offset corresponding to the mcast IP address rewrite
*/
adj->rewrite_header.dst_mcast_offset = offset;
}
static void
fib_entry_src_interface_path_swap (fib_entry_src_t *src,
const fib_entry_t *entry,
fib_path_list_flags_t pl_flags,
const fib_route_path_t *paths)
{
ip_adjacency_t *adj;
src->fes_pl = fib_path_list_create(pl_flags, paths);
/*
* this is a hack to get the entry's prefix into the glean adjacnecy
* so that it is available for fast retreival in the switch path.
*/
if (!(FIB_ENTRY_FLAG_LOCAL & src->fes_entry_flags))
{
adj = adj_get(fib_path_list_get_adj(
src->fes_pl,
fib_entry_get_default_chain_type(entry)));
if (IP_LOOKUP_NEXT_GLEAN == adj->lookup_next_index)
{
/*
* the connected prefix will link to a glean on a non-p2p
* u.interface.
*/
adj->sub_type.glean.receive_addr = entry->fe_prefix.fp_addr;
}
}
}
static clib_error_t *
adj_mcast_interface_state_change (vnet_main_t * vnm,
u32 sw_if_index,
u32 flags)
{
/*
* for each mcast on the interface trigger a walk back to the children
*/
fib_protocol_t proto;
ip_adjacency_t *adj;
for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++)
{
if (sw_if_index >= vec_len(adj_mcasts[proto]) ||
ADJ_INDEX_INVALID == adj_mcasts[proto][sw_if_index])
continue;
adj = adj_get(adj_mcasts[proto][sw_if_index]);
fib_node_back_walk_ctx_t bw_ctx = {
.fnbw_reason = (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP ?
FIB_NODE_BW_REASON_FLAG_INTERFACE_UP :
FIB_NODE_BW_REASON_FLAG_INTERFACE_DOWN),
};
fib_walk_sync(FIB_NODE_TYPE_ADJ, adj_get_index(adj), &bw_ctx);
}
return (NULL);
}
VNET_SW_INTERFACE_ADMIN_UP_DOWN_FUNCTION(adj_mcast_interface_state_change);
/**
* @brief Invoked on each SW interface of a HW interface when the
* HW interface state changes
*/
static walk_rc_t
adj_mcast_hw_sw_interface_state_change (vnet_main_t * vnm,
u32 sw_if_index,
void *arg)
{
adj_mcast_interface_state_change(vnm, sw_if_index, (uword) arg);
return (WALK_CONTINUE);
}
u8*
format_adj_mcast (u8* s, va_list *ap)
{
index_t index = va_arg(*ap, index_t);
CLIB_UNUSED(u32 indent) = va_arg(*ap, u32);
ip_adjacency_t * adj = adj_get(index);
s = format(s, "%U-mcast: ",
format_fib_protocol, adj->ia_nh_proto);
if (adj->rewrite_header.flags & VNET_REWRITE_HAS_FEATURES)
s = format(s, "[features] ");
s = format (s, "%U",
format_vnet_rewrite,
&adj->rewrite_header, sizeof (adj->rewrite_data), 0);
return (s);
}
static clib_error_t *
adj_mcast_interface_delete (vnet_main_t * vnm,
u32 sw_if_index,
u32 is_add)
{
/*
* for each mcast on the interface trigger a walk back to the children
*/
fib_protocol_t proto;
ip_adjacency_t *adj;
if (is_add)
{
/*
* not interested in interface additions. we will not back walk
* to resolve paths through newly added interfaces. Why? The control
* plane should have the brains to add interfaces first, then routes.
* So the case where there are paths with a interface that matches
* one just created is the case where the path resolved through an
* interface that was deleted, and still has not been removed. The
* new interface added, is NO GUARANTEE that the interface being
* added now, even though it may have the same sw_if_index, is the
* same interface that the path needs. So tough!
* If the control plane wants these routes to resolve it needs to
* remove and add them again.
*/
return (NULL);
}
for (proto = FIB_PROTOCOL_IP4; proto <= FIB_PROTOCOL_IP6; proto++)
{
if (sw_if_index >= vec_len(adj_mcasts[proto]) ||
ADJ_INDEX_INVALID == adj_mcasts[proto][sw_if_index])
continue;
adj = adj_get(adj_mcasts[proto][sw_if_index]);
fib_node_back_walk_ctx_t bw_ctx = {
.fnbw_reason = FIB_NODE_BW_REASON_FLAG_INTERFACE_DELETE,
};
fib_walk_sync(FIB_NODE_TYPE_ADJ, adj_get_index(adj), &bw_ctx);
}
return (NULL);
}
/*
* adj_mcast_add_or_lock
*
* The next_hop address here is used for source address selection in the DP.
* The mcast adj is added to an interface's connected prefix, the next-hop
* passed here is the local prefix on the same interface.
*/
adj_index_t
adj_mcast_add_or_lock (fib_protocol_t proto,
vnet_link_t link_type,
u32 sw_if_index)
{
ip_adjacency_t * adj;
vec_validate_init_empty(adj_mcasts[proto], sw_if_index, ADJ_INDEX_INVALID);
if (ADJ_INDEX_INVALID == adj_mcasts[proto][sw_if_index])
{
vnet_main_t *vnm;
vnm = vnet_get_main();
adj = adj_alloc(proto);
adj->lookup_next_index = IP_LOOKUP_NEXT_MCAST;
adj->ia_nh_proto = proto;
adj->ia_link = link_type;
adj_mcasts[proto][sw_if_index] = adj_get_index(adj);
adj_lock(adj_get_index(adj));
vnet_rewrite_init(vnm, sw_if_index, link_type,
adj_get_mcast_node(proto),
vnet_tx_node_index_for_sw_interface(vnm, sw_if_index),
&adj->rewrite_header);
/*
* we need a rewrite where the destination IP address is converted
* to the appropriate link-layer address. This is interface specific.
* So ask the interface to do it.
*/
vnet_update_adjacency_for_sw_interface(vnm, sw_if_index,
adj_get_index(adj));
}
else
{
adj = adj_get(adj_mcasts[proto][sw_if_index]);
adj_lock(adj_get_index(adj));
}
return (adj_get_index(adj));
}
u8*
format_adj_mcast_midchain (u8* s, va_list *ap)
{
index_t index = va_arg(*ap, index_t);
CLIB_UNUSED(u32 indent) = va_arg(*ap, u32);
ip_adjacency_t * adj = adj_get(index);
s = format(s, "%U-mcast-midchain: ",
format_fib_protocol, adj->ia_nh_proto);
s = format (s, "%U",
format_vnet_rewrite,
&adj->rewrite_header,
sizeof (adj->rewrite_data), 0);
s = format (s, "\n%Ustacked-on:\n%U%U",
format_white_space, indent,
format_white_space, indent+2,
format_dpo_id, &adj->sub_type.midchain.next_dpo, indent+2);
return (s);
}
/**
* adj_mcast_midchain_update_rewrite
*
* Update the adjacency's rewrite string. A NULL string implies the
* rewrite is reset (i.e. when ARP/ND entry is gone).
* NB: the adj being updated may be handling traffic in the DP.
*/
void
adj_mcast_midchain_update_rewrite (adj_index_t adj_index,
adj_midchain_fixup_t fixup,
const void *fixup_data,
adj_flags_t flags,
u8 *rewrite,
u8 offset,
u32 mask)
{
ip_adjacency_t *adj;
ASSERT(ADJ_INDEX_INVALID != adj_index);
adj = adj_get(adj_index);
/*
* one time only update. since we don't support changing the tunnel
* src,dst, this is all we need.
*/
ASSERT(adj->lookup_next_index == IP_LOOKUP_NEXT_MCAST);
/*
* tunnels can always provide a rewrite.
*/
ASSERT(NULL != rewrite);
adj_midchain_setup(adj_index, fixup, fixup_data, flags);
/*
* update the adj's rewrite string and build the arc
* from the rewrite node to the interface's TX node
*/
adj_nbr_update_rewrite_internal(adj, IP_LOOKUP_NEXT_MCAST_MIDCHAIN,
adj_get_mcast_node(adj->ia_nh_proto),
vnet_tx_node_index_for_sw_interface(
vnet_get_main(),
adj->rewrite_header.sw_if_index),
rewrite);
adj->rewrite_header.dst_mcast_offset = offset;
}
/**
* @brief The LISP-GPE interface registered function to update, i.e.
* provide an rewrite string for, an adjacency.
*/
void
lisp_gpe_update_adjacency (vnet_main_t * vnm, u32 sw_if_index, adj_index_t ai)
{
const lisp_gpe_tunnel_t *lgt;
lisp_gpe_adjacency_t *ladj;
ip_adjacency_t *adj;
ip_address_t rloc;
vnet_link_t linkt;
adj_flags_t af;
index_t lai;
adj = adj_get (ai);
ip46_address_to_ip_address (&adj->sub_type.nbr.next_hop, &rloc);
/*
* find an existing or create a new adj
*/
lai = lisp_adj_find (&rloc, sw_if_index);
ASSERT (INDEX_INVALID != lai);
ladj = pool_elt_at_index (lisp_adj_pool, lai);
lgt = lisp_gpe_tunnel_get (ladj->tunnel_index);
linkt = adj_get_link_type (ai);
af = ADJ_FLAG_MIDCHAIN_IP_STACK;
if (VNET_LINK_ETHERNET == linkt)
af |= ADJ_FLAG_MIDCHAIN_NO_COUNT;
adj_nbr_midchain_update_rewrite
(ai, lisp_gpe_fixup, NULL, af,
lisp_gpe_tunnel_build_rewrite (lgt, ladj,
lisp_gpe_adj_proto_from_vnet_link_type
(linkt)));
lisp_gpe_adj_stack_one (ladj, ai);
}
always_inline uword
adj_midchain_tx_inline (vlib_main_t * vm,
vlib_node_runtime_t * node,
vlib_frame_t * frame,
int interface_count)
{
u32 * from, * to_next, n_left_from, n_left_to_next;
u32 next_index;
vnet_main_t *vnm = vnet_get_main ();
vnet_interface_main_t *im = &vnm->interface_main;
u32 thread_index = vm->thread_index;
/* Vector of buffer / pkt indices we're supposed to process */
from = vlib_frame_vector_args (frame);
/* Number of buffers / pkts */
n_left_from = frame->n_vectors;
/* Speculatively send the first buffer to the last disposition we used */
next_index = node->cached_next_index;
while (n_left_from > 0)
{
/* set up to enqueue to our disposition with index = next_index */
vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
while (n_left_from >= 8 && n_left_to_next > 4)
{
u32 bi0, adj_index0, next0;
const ip_adjacency_t * adj0;
const dpo_id_t *dpo0;
vlib_buffer_t * b0;
u32 bi1, adj_index1, next1;
const ip_adjacency_t * adj1;
const dpo_id_t *dpo1;
vlib_buffer_t * b1;
u32 bi2, adj_index2, next2;
const ip_adjacency_t * adj2;
const dpo_id_t *dpo2;
vlib_buffer_t * b2;
u32 bi3, adj_index3, next3;
const ip_adjacency_t * adj3;
const dpo_id_t *dpo3;
vlib_buffer_t * b3;
/* Prefetch next iteration. */
{
vlib_buffer_t * p4, * p5;
vlib_buffer_t * p6, * p7;
p4 = vlib_get_buffer (vm, from[4]);
p5 = vlib_get_buffer (vm, from[5]);
p6 = vlib_get_buffer (vm, from[6]);
p7 = vlib_get_buffer (vm, from[7]);
vlib_prefetch_buffer_header (p4, LOAD);
vlib_prefetch_buffer_header (p5, LOAD);
vlib_prefetch_buffer_header (p6, LOAD);
vlib_prefetch_buffer_header (p7, LOAD);
}
bi0 = from[0];
to_next[0] = bi0;
bi1 = from[1];
to_next[1] = bi1;
bi2 = from[2];
to_next[2] = bi2;
bi3 = from[3];
to_next[3] = bi3;
from += 4;
to_next += 4;
n_left_from -= 4;
n_left_to_next -= 4;
b0 = vlib_get_buffer(vm, bi0);
b1 = vlib_get_buffer(vm, bi1);
b2 = vlib_get_buffer(vm, bi2);
b3 = vlib_get_buffer(vm, bi3);
/* Follow the DPO on which the midchain is stacked */
adj_index0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX];
adj_index1 = vnet_buffer(b1)->ip.adj_index[VLIB_TX];
adj_index2 = vnet_buffer(b2)->ip.adj_index[VLIB_TX];
adj_index3 = vnet_buffer(b3)->ip.adj_index[VLIB_TX];
adj0 = adj_get(adj_index0);
adj1 = adj_get(adj_index1);
adj2 = adj_get(adj_index2);
adj3 = adj_get(adj_index3);
dpo0 = &adj0->sub_type.midchain.next_dpo;
dpo1 = &adj1->sub_type.midchain.next_dpo;
dpo2 = &adj2->sub_type.midchain.next_dpo;
dpo3 = &adj3->sub_type.midchain.next_dpo;
next0 = dpo0->dpoi_next_node;
next1 = dpo1->dpoi_next_node;
next2 = dpo2->dpoi_next_node;
next3 = dpo3->dpoi_next_node;
vnet_buffer(b1)->ip.adj_index[VLIB_TX] = dpo1->dpoi_index;
vnet_buffer(b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index;
vnet_buffer(b2)->ip.adj_index[VLIB_TX] = dpo2->dpoi_index;
vnet_buffer(b3)->ip.adj_index[VLIB_TX] = dpo3->dpoi_index;
if (interface_count)
{
vlib_increment_combined_counter (im->combined_sw_if_counters
+ VNET_INTERFACE_COUNTER_TX,
thread_index,
adj0->rewrite_header.sw_if_index,
1,
vlib_buffer_length_in_chain (vm, b0));
vlib_increment_combined_counter (im->combined_sw_if_counters
+ VNET_INTERFACE_COUNTER_TX,
thread_index,
adj1->rewrite_header.sw_if_index,
1,
vlib_buffer_length_in_chain (vm, b1));
vlib_increment_combined_counter (im->combined_sw_if_counters
+ VNET_INTERFACE_COUNTER_TX,
thread_index,
adj2->rewrite_header.sw_if_index,
1,
vlib_buffer_length_in_chain (vm, b2));
vlib_increment_combined_counter (im->combined_sw_if_counters
+ VNET_INTERFACE_COUNTER_TX,
thread_index,
adj3->rewrite_header.sw_if_index,
1,
vlib_buffer_length_in_chain (vm, b3));
}
if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
{
adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node,
b0, sizeof (*tr));
tr->ai = adj_index0;
}
if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED))
{
adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node,
b1, sizeof (*tr));
tr->ai = adj_index1;
}
if (PREDICT_FALSE(b2->flags & VLIB_BUFFER_IS_TRACED))
{
adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node,
b2, sizeof (*tr));
tr->ai = adj_index2;
}
if (PREDICT_FALSE(b3->flags & VLIB_BUFFER_IS_TRACED))
{
adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node,
b3, sizeof (*tr));
tr->ai = adj_index3;
}
vlib_validate_buffer_enqueue_x4 (vm, node, next_index,
to_next, n_left_to_next,
bi0, bi1, bi2, bi3,
next0, next1, next2, next3);
}
while (n_left_from > 0 && n_left_to_next > 0)
{
u32 bi0, adj_index0, next0;
const ip_adjacency_t * adj0;
const dpo_id_t *dpo0;
vlib_buffer_t * b0;
bi0 = from[0];
to_next[0] = bi0;
from += 1;
to_next += 1;
n_left_from -= 1;
n_left_to_next -= 1;
b0 = vlib_get_buffer(vm, bi0);
/* Follow the DPO on which the midchain is stacked */
adj_index0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX];
adj0 = adj_get(adj_index0);
dpo0 = &adj0->sub_type.midchain.next_dpo;
next0 = dpo0->dpoi_next_node;
vnet_buffer(b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index;
if (interface_count)
{
vlib_increment_combined_counter (im->combined_sw_if_counters
+ VNET_INTERFACE_COUNTER_TX,
thread_index,
adj0->rewrite_header.sw_if_index,
1,
vlib_buffer_length_in_chain (vm, b0));
}
if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
{
adj_midchain_tx_trace_t *tr = vlib_add_trace (vm, node,
b0, sizeof (*tr));
tr->ai = adj_index0;
}
vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
to_next, n_left_to_next,
bi0, next0);
}
vlib_put_next_frame (vm, node, next_index, n_left_to_next);
}
return frame->n_vectors;
}
