summaryrefslogtreecommitdiffstats
path: root/vnet
diff options
context:
space:
mode:
authorDave Barach <dave@barachs.net>2016-10-18 15:25:35 -0400
committerDamjan Marion <dmarion.lists@gmail.com>2016-10-21 20:16:40 +0000
commit670909eb6eea7c0d0c0e8d674771db23bb018f4a (patch)
tree47c421c0ee1162222ecd65f310e726a36dcf0bd0 /vnet
parent7b0196b8262a7a441024514976fcc29980f3fe6c (diff)
Quad-loop ip4_lookup_inline, add _x4 validate buffer enqueue
Change-Id: I80a25fa90d8a65db619b8697cede2b98bed37cea Signed-off-by: Dave Barach <dave@barachs.net>
Diffstat (limited to 'vnet')
-rw-r--r--vnet/vnet/ip/ip4_forward.c180
1 files changed, 115 insertions, 65 deletions
diff --git a/vnet/vnet/ip/ip4_forward.c b/vnet/vnet/ip/ip4_forward.c
index 0ad87289519..9e712f5493a 100644
--- a/vnet/vnet/ip/ip4_forward.c
+++ b/vnet/vnet/ip/ip4_forward.c
@@ -85,115 +85,167 @@ ip4_lookup_inline (vlib_main_t * vm,
vlib_get_next_frame (vm, node, next,
to_next, n_left_to_next);
- while (n_left_from >= 4 && n_left_to_next >= 2)
+ while (n_left_from >= 8 && n_left_to_next >= 4)
{
- vlib_buffer_t * p0, * p1;
- ip4_header_t * ip0, * ip1;
- __attribute__((unused)) tcp_header_t * tcp0, * tcp1;
- ip_lookup_next_t next0, next1;
- const load_balance_t * lb0, * lb1;
- ip4_fib_mtrie_t * mtrie0, * mtrie1;
- ip4_fib_mtrie_leaf_t leaf0, leaf1;
- ip4_address_t * dst_addr0, *dst_addr1;
+ vlib_buffer_t * p0, * p1, * p2, * p3;
+ ip4_header_t * ip0, * ip1, * ip2, * ip3;
+ __attribute__((unused)) tcp_header_t * tcp0, * tcp1, * tcp2, * tcp3;
+ ip_lookup_next_t next0, next1, next2, next3;
+ const load_balance_t * lb0, * lb1, * lb2, * lb3;
+ ip4_fib_mtrie_t * mtrie0, * mtrie1, * mtrie2, * mtrie3;
+ ip4_fib_mtrie_leaf_t leaf0, leaf1, leaf2, leaf3;
+ ip4_address_t * dst_addr0, *dst_addr1, *dst_addr2, *dst_addr3;
__attribute__((unused)) u32 pi0, fib_index0, lb_index0, is_tcp_udp0;
__attribute__((unused)) u32 pi1, fib_index1, lb_index1, is_tcp_udp1;
+ __attribute__((unused)) u32 pi2, fib_index2, lb_index2, is_tcp_udp2;
+ __attribute__((unused)) u32 pi3, fib_index3, lb_index3, is_tcp_udp3;
flow_hash_config_t flow_hash_config0, flow_hash_config1;
- u32 hash_c0, hash_c1;
- u32 wrong_next;
- const dpo_id_t *dpo0, *dpo1;
+ flow_hash_config_t flow_hash_config2, flow_hash_config3;
+ u32 hash_c0, hash_c1, hash_c2, hash_c3;
+ const dpo_id_t *dpo0, *dpo1, *dpo2, *dpo3;
/* Prefetch next iteration. */
{
- vlib_buffer_t * p2, * p3;
-
- p2 = vlib_get_buffer (vm, from[2]);
- p3 = vlib_get_buffer (vm, from[3]);
-
- vlib_prefetch_buffer_header (p2, LOAD);
- vlib_prefetch_buffer_header (p3, LOAD);
-
- CLIB_PREFETCH (p2->data, sizeof (ip0[0]), LOAD);
- CLIB_PREFETCH (p3->data, sizeof (ip0[0]), LOAD);
+ vlib_buffer_t * p4, * p5, * 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);
+
+ CLIB_PREFETCH (p4->data, sizeof (ip0[0]), LOAD);
+ CLIB_PREFETCH (p5->data, sizeof (ip0[0]), LOAD);
+ CLIB_PREFETCH (p6->data, sizeof (ip0[0]), LOAD);
+ CLIB_PREFETCH (p7->data, sizeof (ip0[0]), LOAD);
}
pi0 = to_next[0] = from[0];
pi1 = to_next[1] = from[1];
+ pi2 = to_next[2] = from[2];
+ pi3 = to_next[3] = from[3];
+
+ from += 4;
+ to_next += 4;
+ n_left_to_next -= 4;
+ n_left_from -= 4;
p0 = vlib_get_buffer (vm, pi0);
p1 = vlib_get_buffer (vm, pi1);
+ p2 = vlib_get_buffer (vm, pi2);
+ p3 = vlib_get_buffer (vm, pi3);
ip0 = vlib_buffer_get_current (p0);
ip1 = vlib_buffer_get_current (p1);
+ ip2 = vlib_buffer_get_current (p2);
+ ip3 = vlib_buffer_get_current (p3);
dst_addr0 = &ip0->dst_address;
dst_addr1 = &ip1->dst_address;
+ dst_addr2 = &ip2->dst_address;
+ dst_addr3 = &ip3->dst_address;
fib_index0 = vec_elt (im->fib_index_by_sw_if_index, vnet_buffer (p0)->sw_if_index[VLIB_RX]);
fib_index1 = vec_elt (im->fib_index_by_sw_if_index, vnet_buffer (p1)->sw_if_index[VLIB_RX]);
+ fib_index2 = vec_elt (im->fib_index_by_sw_if_index, vnet_buffer (p2)->sw_if_index[VLIB_RX]);
+ fib_index3 = vec_elt (im->fib_index_by_sw_if_index, vnet_buffer (p3)->sw_if_index[VLIB_RX]);
fib_index0 = (vnet_buffer(p0)->sw_if_index[VLIB_TX] == (u32)~0) ?
fib_index0 : vnet_buffer(p0)->sw_if_index[VLIB_TX];
fib_index1 = (vnet_buffer(p1)->sw_if_index[VLIB_TX] == (u32)~0) ?
fib_index1 : vnet_buffer(p1)->sw_if_index[VLIB_TX];
+ fib_index2 = (vnet_buffer(p2)->sw_if_index[VLIB_TX] == (u32)~0) ?
+ fib_index2 : vnet_buffer(p2)->sw_if_index[VLIB_TX];
+ fib_index3 = (vnet_buffer(p3)->sw_if_index[VLIB_TX] == (u32)~0) ?
+ fib_index3 : vnet_buffer(p3)->sw_if_index[VLIB_TX];
if (! lookup_for_responses_to_locally_received_packets)
{
mtrie0 = &ip4_fib_get (fib_index0)->mtrie;
mtrie1 = &ip4_fib_get (fib_index1)->mtrie;
+ mtrie2 = &ip4_fib_get (fib_index2)->mtrie;
+ mtrie3 = &ip4_fib_get (fib_index3)->mtrie;
- leaf0 = leaf1 = IP4_FIB_MTRIE_LEAF_ROOT;
+ leaf0 = leaf1 = leaf2 = leaf3 = IP4_FIB_MTRIE_LEAF_ROOT;
leaf0 = ip4_fib_mtrie_lookup_step (mtrie0, leaf0, dst_addr0, 0);
leaf1 = ip4_fib_mtrie_lookup_step (mtrie1, leaf1, dst_addr1, 0);
+ leaf2 = ip4_fib_mtrie_lookup_step (mtrie2, leaf2, dst_addr2, 0);
+ leaf3 = ip4_fib_mtrie_lookup_step (mtrie3, leaf3, dst_addr3, 0);
}
tcp0 = (void *) (ip0 + 1);
tcp1 = (void *) (ip1 + 1);
+ tcp2 = (void *) (ip2 + 1);
+ tcp3 = (void *) (ip3 + 1);
is_tcp_udp0 = (ip0->protocol == IP_PROTOCOL_TCP
|| ip0->protocol == IP_PROTOCOL_UDP);
is_tcp_udp1 = (ip1->protocol == IP_PROTOCOL_TCP
|| ip1->protocol == IP_PROTOCOL_UDP);
+ is_tcp_udp2 = (ip2->protocol == IP_PROTOCOL_TCP
+ || ip2->protocol == IP_PROTOCOL_UDP);
+ is_tcp_udp3 = (ip1->protocol == IP_PROTOCOL_TCP
+ || ip1->protocol == IP_PROTOCOL_UDP);
if (! lookup_for_responses_to_locally_received_packets)
{
leaf0 = ip4_fib_mtrie_lookup_step (mtrie0, leaf0, dst_addr0, 1);
leaf1 = ip4_fib_mtrie_lookup_step (mtrie1, leaf1, dst_addr1, 1);
+ leaf2 = ip4_fib_mtrie_lookup_step (mtrie2, leaf2, dst_addr2, 1);
+ leaf3 = ip4_fib_mtrie_lookup_step (mtrie3, leaf3, dst_addr3, 1);
}
if (! lookup_for_responses_to_locally_received_packets)
{
leaf0 = ip4_fib_mtrie_lookup_step (mtrie0, leaf0, dst_addr0, 2);
leaf1 = ip4_fib_mtrie_lookup_step (mtrie1, leaf1, dst_addr1, 2);
+ leaf2 = ip4_fib_mtrie_lookup_step (mtrie2, leaf2, dst_addr2, 2);
+ leaf3 = ip4_fib_mtrie_lookup_step (mtrie3, leaf3, dst_addr3, 2);
}
if (! lookup_for_responses_to_locally_received_packets)
{
leaf0 = ip4_fib_mtrie_lookup_step (mtrie0, leaf0, dst_addr0, 3);
leaf1 = ip4_fib_mtrie_lookup_step (mtrie1, leaf1, dst_addr1, 3);
+ leaf2 = ip4_fib_mtrie_lookup_step (mtrie2, leaf2, dst_addr2, 3);
+ leaf3 = ip4_fib_mtrie_lookup_step (mtrie3, leaf3, dst_addr3, 3);
}
if (lookup_for_responses_to_locally_received_packets)
{
lb_index0 = vnet_buffer (p0)->ip.adj_index[VLIB_RX];
lb_index1 = vnet_buffer (p1)->ip.adj_index[VLIB_RX];
+ lb_index2 = vnet_buffer (p2)->ip.adj_index[VLIB_RX];
+ lb_index3 = vnet_buffer (p3)->ip.adj_index[VLIB_RX];
}
else
{
/* Handle default route. */
leaf0 = (leaf0 == IP4_FIB_MTRIE_LEAF_EMPTY ? mtrie0->default_leaf : leaf0);
leaf1 = (leaf1 == IP4_FIB_MTRIE_LEAF_EMPTY ? mtrie1->default_leaf : leaf1);
-
+ leaf2 = (leaf2 == IP4_FIB_MTRIE_LEAF_EMPTY ? mtrie2->default_leaf : leaf2);
+ leaf3 = (leaf3 == IP4_FIB_MTRIE_LEAF_EMPTY ? mtrie3->default_leaf : leaf3);
lb_index0 = ip4_fib_mtrie_leaf_get_adj_index (leaf0);
lb_index1 = ip4_fib_mtrie_leaf_get_adj_index (leaf1);
+ lb_index2 = ip4_fib_mtrie_leaf_get_adj_index (leaf2);
+ lb_index3 = ip4_fib_mtrie_leaf_get_adj_index (leaf3);
}
lb0 = load_balance_get (lb_index0);
lb1 = load_balance_get (lb_index1);
+ lb2 = load_balance_get (lb_index2);
+ lb3 = load_balance_get (lb_index3);
/* Use flow hash to compute multipath adjacency. */
hash_c0 = vnet_buffer (p0)->ip.flow_hash = 0;
hash_c1 = vnet_buffer (p1)->ip.flow_hash = 0;
+ hash_c2 = vnet_buffer (p2)->ip.flow_hash = 0;
+ hash_c3 = vnet_buffer (p3)->ip.flow_hash = 0;
if (PREDICT_FALSE (lb0->lb_n_buckets > 1))
{
flow_hash_config0 = lb0->lb_hash_config;
@@ -206,11 +258,27 @@ ip4_lookup_inline (vlib_main_t * vm,
hash_c1 = vnet_buffer (p1)->ip.flow_hash =
ip4_compute_flow_hash (ip1, flow_hash_config1);
}
+ if (PREDICT_FALSE (lb2->lb_n_buckets > 1))
+ {
+ flow_hash_config2 = lb2->lb_hash_config;
+ hash_c2 = vnet_buffer (p2)->ip.flow_hash =
+ ip4_compute_flow_hash (ip2, flow_hash_config2);
+ }
+ if (PREDICT_FALSE(lb3->lb_n_buckets > 1))
+ {
+ flow_hash_config3 = lb3->lb_hash_config;
+ hash_c3 = vnet_buffer (p3)->ip.flow_hash =
+ ip4_compute_flow_hash (ip3, flow_hash_config3);
+ }
ASSERT (lb0->lb_n_buckets > 0);
ASSERT (is_pow2 (lb0->lb_n_buckets));
ASSERT (lb1->lb_n_buckets > 0);
ASSERT (is_pow2 (lb1->lb_n_buckets));
+ ASSERT (lb2->lb_n_buckets > 0);
+ ASSERT (is_pow2 (lb2->lb_n_buckets));
+ ASSERT (lb3->lb_n_buckets > 0);
+ ASSERT (is_pow2 (lb3->lb_n_buckets));
dpo0 = load_balance_get_bucket_i(lb0,
(hash_c0 &
@@ -218,11 +286,21 @@ ip4_lookup_inline (vlib_main_t * vm,
dpo1 = load_balance_get_bucket_i(lb1,
(hash_c1 &
(lb1->lb_n_buckets_minus_1)));
+ dpo2 = load_balance_get_bucket_i(lb2,
+ (hash_c2 &
+ (lb2->lb_n_buckets_minus_1)));
+ dpo3 = load_balance_get_bucket_i(lb3,
+ (hash_c3 &
+ (lb3->lb_n_buckets_minus_1)));
next0 = dpo0->dpoi_next_node;
vnet_buffer (p0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index;
next1 = dpo1->dpoi_next_node;
vnet_buffer (p1)->ip.adj_index[VLIB_TX] = dpo1->dpoi_index;
+ next2 = dpo2->dpoi_next_node;
+ vnet_buffer (p2)->ip.adj_index[VLIB_TX] = dpo2->dpoi_index;
+ next3 = dpo3->dpoi_next_node;
+ vnet_buffer (p3)->ip.adj_index[VLIB_TX] = dpo3->dpoi_index;
vlib_increment_combined_counter
(cm, cpu_index, lb_index0, 1,
@@ -232,48 +310,20 @@ ip4_lookup_inline (vlib_main_t * vm,
(cm, cpu_index, lb_index1, 1,
vlib_buffer_length_in_chain (vm, p1)
+ sizeof(ethernet_header_t));
+ vlib_increment_combined_counter
+ (cm, cpu_index, lb_index2, 1,
+ vlib_buffer_length_in_chain (vm, p2)
+ + sizeof(ethernet_header_t));
+ vlib_increment_combined_counter
+ (cm, cpu_index, lb_index3, 1,
+ vlib_buffer_length_in_chain (vm, p3)
+ + sizeof(ethernet_header_t));
- from += 2;
- to_next += 2;
- n_left_to_next -= 2;
- n_left_from -= 2;
-
- wrong_next = (next0 != next) + 2*(next1 != next);
- if (PREDICT_FALSE (wrong_next != 0))
- {
- switch (wrong_next)
- {
- case 1:
- /* A B A */
- to_next[-2] = pi1;
- to_next -= 1;
- n_left_to_next += 1;
- vlib_set_next_frame_buffer (vm, node, next0, pi0);
- break;
-
- case 2:
- /* A A B */
- to_next -= 1;
- n_left_to_next += 1;
- vlib_set_next_frame_buffer (vm, node, next1, pi1);
- break;
-
- case 3:
- /* A B C */
- to_next -= 2;
- n_left_to_next += 2;
- vlib_set_next_frame_buffer (vm, node, next0, pi0);
- vlib_set_next_frame_buffer (vm, node, next1, pi1);
- if (next0 == next1)
- {
- /* A B B */
- vlib_put_next_frame (vm, node, next, n_left_to_next);
- next = next1;
- vlib_get_next_frame (vm, node, next, to_next, n_left_to_next);
- }
- }
- }
- }
+ vlib_validate_buffer_enqueue_x4 (vm, node, next,
+ to_next, n_left_to_next,
+ pi0, pi1, pi2, pi3,
+ next0, next1, next2, next3);
+ }
while (n_left_from > 0 && n_left_to_next > 0)
{
s> peer-addr <address> desired-min-tx <interval> required-min-rx <interval> detect-mult <multiplier> Parameters: * interface - interface to which this session is tied to * local-addr - local address (ipv4 or ipv6) * peer-addr - peer address (ipv4 or ipv6, must match local-addr family) * desired-min-tx - desired minimum tx interval (microseconds) * required-min-rx - required minimum rx interval (microseconds) * detect-mult - detect multiplier (must be non-zero) Example: > bfd udp session mod interface pg0 local-addr 172.16.1.1 peer-addr 172.16.1.2 desired-min-tx 300000 required-min-rx 200000 detect-mult 12 Notes: * desired-min-tx controls desired transmission rate of both control frames and echo packets ##### Delete an existing BFD session > bfd udp session del interface <interface> local-addr <address> peer-addr<address> Parameters: * interface - interface to which this session is tied to * local-addr - local address (ipv4 or ipv6) * peer-addr - peer address (ipv4 or ipv6, must match local-addr family) Example: > bfd udp session del interface pg0 local-addr 172.16.1.1 peer-addr 172.16.1.2 ##### Set session admin-up or admin-down > bfd udp session set-flags interface <interface> local-addr <address> peer-addr <address> admin <up|down> Parameters: * interface - interface to which this session is tied to * local-addr - local address (ipv4 or ipv6) * peer-addr - peer address (ipv4 or ipv6, must match local-addr family) * admin - up/down based on desired action Example: > bfd udp session set-flags admin down interface pg0 local-addr 172.16.1.1 peer-addr 172.16.1.2 ##### Activate/change authentication for existing session > bfd udp session auth activate interface <interface> local-addr <address> peer-addr <address> conf-key-id <ID> bfd-key-id <ID> [ delayed <yes|no> ] Parameters: * interface - interface to which this session is tied to * local-addr - local address (ipv4 or ipv6) * peer-addr - peer address (ipv4 or ipv6, must match local-addr family) * conf-key-id - local configuration key ID * bfd-key-id - BFD key ID, as carried in BFD control frames * delayed - is yes then this action is delayed until the peer performs the same action Example: > bfd udp session auth activate interface pg0 local-addr 172.16.1.1 peer-addr 172.16.1.2 conf-key-id 540928695 bfd-key-id 239 delayed yes Notes: * see [Delayed option] for more information ##### Deactivate authentication for existing session > bfd udp session auth deactivate interface <interface> local-addr <address> peer-addr <address> [ delayed <yes|no> ] Parameters: * interface - interface to which this session is tied to * local-addr - local address (ipv4 or ipv6) * peer-addr - peer address (ipv4 or ipv6, must match local-addr family) * delayed - is yes then this action is delayed until the peer performs the same action Example: > bfd udp session auth deactivate interface pg0 local-addr 172.16.1.1 peer-addr 172.16.1.2 Notes: * see [Delayed option] for more information ##### Set echo-source interface > bfd udp echo-source set interface <interface> Parameters: * interface - interface used for getting source address for echo packets Example: > bfd udp echo-source set interface loop0 ##### Delete echo-source interface > bfd udp echo-source del Example: > bfd udp echo-source del ### Authentication BFD sessions should be authenticated for security purposes. SHA1 and meticulous SHA1 authentication is supported by VPP. First, authentication keys are configured in VPP and afterwards they can be used by sessions. There are two key IDs in the scope of BFD session: * configuration key ID is the internal unique key ID inside VPP and is never communicated to any peer, it serves only the purpose of identifying the key * BFD key ID is the key ID carried in BFD control frames and is used for verifying authentication #### Turning auth on/off Authentication can be turned on or off at any time. Care must be taken however, to either synchronize the authentication manipulation with peer's actions to avoid the session going down. ##### Delayed option Delayed option is useful for synchronizing authentication changes with a peer. If it's specified, then authentication change is not performed immediately. In this case, VPP continues to transmit packets using the old authentication method (unauthenticated or using old sha1 key). If a packet is received, which does not pass the current authentication, then VPP tries to authenticate it using the new method (which might be none, if deactivating authentication) and if it passes, then the new authentication method is put in use. The recommended procedure for enabling/changing/disabling session authentication is: 1. perform authentication change on vpp's side with delayed option set to yes 2. perform authentication change on peer's side (without delayed option) Notes: * if both peers use delayed option at the same time, the change will never be carried out, since none of the peers will see any packet with the new authentication which could trigger the change * remote peer does not need to support or even be aware of this mechanism for it to work properly ### Echo function Echo function is used by VPP whenever a peer declares the willingness to support it, echo-source is set and it contains a usable subnet (see below). When echo function is switched on, the required min rx interval advertised to peer is set to 1 second (or the configured value, if its higher). #### Echo source address Because echo packets are only looped back (and not processed in any way) by a peer, it's necessary to set the source address in a way which avoids packet drop due to spoofing protection by VPP. Per RFC, the source address should not be in the subnet set on the interface over which the echo packets are sent. Also, it must not be any VPP-local address, otherwise the packet gets dropped on receipt by VPP. The solution is to create a loopback interface with a (private) IPv4/IPv6 subnet assigned as echo-source. The BFD then picks an unused address from the subnet by flipping the last bit and uses that as source address in the echo packets, thus meeting RFC recommendation while avoiding spoofing protection. Example: if 10.10.10.3/31 is the subnet, then 10.10.10.2 will be used as source address in (IPv4) echo packets ### Demand mode Demand mode is respected by VPP, but not used locally. The only scenario when demand mode could make sense currently is when echo is active. Because echo packets are inherently insecure against an adversary looping them back a poll sequence would be required for slow periodic connectivity verification anyway. It's more efficient to just ask the remote peer to send slow periodic control frames without VPP initiating periodic poll sequences. ### Admin-down Session may be put admin-down at any time. This immediately causes the state to be changed to AdminDown and remain so unless the session is put admin-up. ## BFD implementation notes Because BFD can work over different transport layers, the BFD code is separated into core BFD functionality - main module implemented in bfd_main.c and transport-specific code implemented in bfd_udp.c. ### Main module Main module is responsible for handling all the BFD functionality defined in RFC 5880. #### Internal API Internal APIs defined in bfd_main.h are called from transport-specific code to create/modify/delete #### Packet receipt When a packet is received by the transport layer, it is forwarded to main module (to main thread) via an RPC call. At this point, the authentication has been verified, so the packet is consumed, session parameters are updated accordingly and state change (if applicable). Based on these, the timeouts are adjusted if required and an event is sent to the process node to wake up and recalculate sleep time. #### Packet transmit Main module allocates a vlib_buffer_t, creates the required BFD frame (control or echo in it), then calls the transport layer to add the transport layer. Then a frame containing the buffer to the aprropriate node is created and enqueued. #### Process node Main module implements one process node which is a simple loop. The process node gets next timeout from the timer wheel, sleeps until the timeout expires and then calls a timeout routine which drives the state machine for each session which timed out. The sleep is interrupted externally via vlib event, when a session is added or modified in a way which might require timer wheel manipulation. In this case the caller inserts the necessary timeout to timer wheel and then signals the process node to wake up early, handle possible timeouts and recalculate the sleep time again. #### State machine Default state of BFD session when created is Down, per RFC 5880. State changes to Init, Up or Down based on events like received state from peer and timeouts. The session state can be set AdminDown using a binary API, which prevents it from going to any other state, until this limitation is removed. This state is advertised to peers in slow periodic control frames. For each session, the following timeouts are maintained: 1. tx timeout - used for sending out control frames 2. rx timeout - used for detecting session timeout 3. echo tx timeout - used for sending out echo frames 3. echo rx timeout - used for detecting session timeout based on echo These timeouts are maintained in cpu clocks and recalculated when appropriate (e.g. rx timeout is bumped when a packet is received, keeping the session alive). Only the earliest timeout is inserted into the timer wheel at a time and timer wheel events are never deleted, rather spurious events are ignored. This allows efficient operation, like not inserting events into timing wheel for each packet received or ignoring left-over events in case a bfd session gets removed and a new one is recreated with the same session index. #### Authentication keys management Authentication keys are managed internally in a pool, with each key tracking it's use count. The removal/modification is only allowed if the key is not in use. ### UDP module UDP module is responsible for: 1. public APIs/CLIs to configure BFD over UDP. 2. support code called by main module to encapsulate/decapsulate BFD packets This module implements two graph nodes - for consuming ipv4 and ipv6 packets target at BFD ports 3874 and 3875. #### Packet receipt BFD packet receipt receipt starts in the bfd udp graph nodes. Since the code needs to verify IP/UDP header data, it relies on ip4-local (and ip6-local) nodes to store pointers to the appropriate headers. First, your discriminator is extracted from BFD packet and used to lookup the existing session. In case it's zero, the pair of IP addresses and sw_if_index is used to lookup session. Then, main module is called to verify the authentication, if present. Afterwards a check is made if the IP/UDP headers are correct. If yes, then an RPC call is made to the main thread to consume the packet and take action upon it. #### Packet transmission When process node decides that there is a need to transmit the packet, it creates a buffer, fills the BFD frame data in and calls the UDP module to add the transport layer. This is a simple operation for the control frames consisting of just adding UDP/IP headers based on session data. For echo frames, an additional step, looking at the echo-source interface and picking and address is performed and if this fails, then the packet cannot be transmitted and an error is returned to main thread.