static inline int
session_send_evt_to_thread (void *data, void *args, u32 thread_index,
session_evt_type_t evt_type)
{
session_event_t *evt;
svm_msg_q_msg_t msg;
svm_msg_q_t *mq;
u32 tries = 0, max_tries;
mq = session_main_get_vpp_event_queue (thread_index);
while (svm_msg_q_try_lock (mq))
{
max_tries = vlib_get_current_process (vlib_get_main ())? 1e6 : 3;
if (tries++ == max_tries)
{
SESSION_DBG ("failed to enqueue evt");
return -1;
}
}
if (PREDICT_FALSE (svm_msg_q_ring_is_full (mq, SESSION_MQ_IO_EVT_RING)))
{
svm_msg_q_unlock (mq);
return -2;
}
msg = svm_msg_q_alloc_msg_w_ring (mq, SESSION_MQ_IO_EVT_RING);
if (PREDICT_FALSE (svm_msg_q_msg_is_invalid (&msg)))
{
svm_msg_q_unlock (mq);
return -2;
}
evt = (session_event_t *) svm_msg_q_msg_data (mq, &msg);
evt->event_type = evt_type;
switch (evt_type)
{
case SESSION_CTRL_EVT_RPC:
evt->rpc_args.fp = data;
evt->rpc_args.arg = args;
break;
case SESSION_IO_EVT_TX:
case SESSION_IO_EVT_TX_FLUSH:
case SESSION_IO_EVT_BUILTIN_RX:
evt->session_index = *(u32 *) data;
break;
case SESSION_IO_EVT_BUILTIN_TX:
case SESSION_CTRL_EVT_CLOSE:
evt->session_handle = session_handle ((session_t *) data);
break;
default:
clib_warning ("evt unhandled!");
svm_msg_q_unlock (mq);
return -1;
}
svm_msg_q_add_and_unlock (mq, &msg);
return 0;
}
static void init_keys_direct_u32 (phash_main_t * pm)
{
int n_keys_left, b_mask, a_shift;
u32 seed;
phash_key_t * k;
seed = pm->hash_seed;
b_mask = (1 << pm->b_bits) - 1;
a_shift = BITS (seed) - pm->a_bits;
k = pm->keys;
n_keys_left = vec_len (pm->keys);
while (n_keys_left >= 2)
{
u32 x0, y0, z0;
u32 x1, y1, z1;
x0 = y0 = z0 = seed;
x1 = y1 = z1 = seed;
x0 += (u32) k[0].key;
x1 += (u32) k[1].key;
hash_mix32 (x0, y0, z0);
hash_mix32 (x1, y1, z1);
k[0].b = z0 & b_mask;
k[1].b = z1 & b_mask;
k[0].a = z0 >> a_shift;
k[1].a = z1 >> a_shift;
if (PREDICT_FALSE (a_shift >= BITS (z0)))
k[0].a = k[1].a = 0;
k += 2;
n_keys_left -= 2;
}
if (n_keys_left >= 1)
{
u32 x0, y0, z0;
x0 = y0 = z0 = seed;
x0 += k[0].key;
hash_mix32 (x0, y0, z0);
k[0].b = z0 & b_mask;
k[0].a = z0 >> a_shift;
if (PREDICT_FALSE (a_shift >= BITS (z0)))
k[0].a = 0;
k += 1;
n_keys_left -= 1;
}
/*
* Enqueue data for delivery to session peer. Does not notify peer of enqueue
* event but on request can queue notification events for later delivery by
* calling stream_server_flush_enqueue_events().
*
* @param tc Transport connection which is to be enqueued data
* @param b Buffer to be enqueued
* @param offset Offset at which to start enqueueing if out-of-order
* @param queue_event Flag to indicate if peer is to be notified or if event
* is to be queued. The former is useful when more data is
* enqueued and only one event is to be generated.
* @param is_in_order Flag to indicate if data is in order
* @return Number of bytes enqueued or a negative value if enqueueing failed.
*/
int
session_enqueue_stream_connection (transport_connection_t * tc,
vlib_buffer_t * b, u32 offset,
u8 queue_event, u8 is_in_order)
{
session_t *s;
int enqueued = 0, rv, in_order_off;
s = session_get (tc->s_index, tc->thread_index);
if (is_in_order)
{
enqueued = svm_fifo_enqueue (s->rx_fifo,
b->current_length,
vlib_buffer_get_current (b));
if (PREDICT_FALSE ((b->flags & VLIB_BUFFER_NEXT_PRESENT)
&& enqueued >= 0))
{
in_order_off = enqueued > b->current_length ? enqueued : 0;
rv = session_enqueue_chain_tail (s, b, in_order_off, 1);
if (rv > 0)
enqueued += rv;
}
}
else
{
rv = svm_fifo_enqueue_with_offset (s->rx_fifo, offset,
b->current_length,
vlib_buffer_get_current (b));
if (PREDICT_FALSE ((b->flags & VLIB_BUFFER_NEXT_PRESENT) && !rv))
session_enqueue_chain_tail (s, b, offset + b->current_length, 0);
/* if something was enqueued, report even this as success for ooo
* segment handling */
return rv;
}
if (queue_event)
{
/* Queue RX event on this fifo. Eventually these will need to be flushed
* by calling stream_server_flush_enqueue_events () */
session_worker_t *wrk;
wrk = session_main_get_worker (s->thread_index);
if (!(s->flags & SESSION_F_RX_EVT))
{
s->flags |= SESSION_F_RX_EVT;
vec_add1 (wrk->session_to_enqueue[tc->proto], s->session_index);
}
}
return enqueued;
}
static struct rte_mbuf * dpdk_replicate_packet_mb (vlib_buffer_t * b)
{
vlib_main_t * vm = vlib_get_main();
vlib_buffer_main_t * bm = vm->buffer_main;
struct rte_mbuf * first_mb = 0, * new_mb, * pkt_mb, ** prev_mb_next = 0;
u8 nb_segs, nb_segs_left;
u32 copy_bytes;
unsigned socket_id = rte_socket_id();
ASSERT (bm->pktmbuf_pools[socket_id]);
pkt_mb = ((struct rte_mbuf *)b)-1;
nb_segs = pkt_mb->nb_segs;
for (nb_segs_left = nb_segs; nb_segs_left; nb_segs_left--)
{
if (PREDICT_FALSE(pkt_mb == 0))
{
clib_warning ("Missing %d mbuf chain segment(s): "
"(nb_segs = %d, nb_segs_left = %d)!",
nb_segs - nb_segs_left, nb_segs, nb_segs_left);
if (first_mb)
rte_pktmbuf_free(first_mb);
return NULL;
}
new_mb = rte_pktmbuf_alloc (bm->pktmbuf_pools[socket_id]);
if (PREDICT_FALSE(new_mb == 0))
{
if (first_mb)
rte_pktmbuf_free(first_mb);
return NULL;
}
/*
* Copy packet info into 1st segment.
*/
if (first_mb == 0)
{
first_mb = new_mb;
rte_pktmbuf_pkt_len (first_mb) = pkt_mb->pkt_len;
first_mb->nb_segs = pkt_mb->nb_segs;
first_mb->port = pkt_mb->port;
#ifdef DAW_FIXME // TX Offload support TBD
first_mb->vlan_macip = pkt_mb->vlan_macip;
first_mb->hash = pkt_mb->hash;
first_mb->ol_flags = pkt_mb->ol_flags
#endif
}
else
{
void cnat_port_free_v2 (
cnat_portmap_v2_t *pm,
int index,
port_pair_t pair_type,
u16 base_port,
u16 static_port_range)
{
cnat_portmap_v2_t *my_pm;
uword bit;
/* check for valid portmap */
if (PREDICT_FALSE(index > vec_len(pm))) {
spp_printf(CNAT_INVALID_INDEX_TO_FREE_PORT, 0, 0);
return;
}
my_pm = pm + index;
bit = port2bit(base_port);
#if DEBUG > 0
if(clib_bitmap_get_no_check(my_pm->bm, bit))
ASSERT(clib_bitmap_get_no_check(my_pm->bm, bit) == 0);
#endif
cgn_clib_bitmap_set_no_check(my_pm->bm, bit);
my_pm->inuse -= 1;
if(base_port >= static_port_range) {
/* Clear the full flag. we can have a new dynamic session now */
my_pm->dyn_full = 0;
}
return;
}
unsigned int CompressedSet::Iterator::nextDoc(){
//: if the pointer points to the end
if(PREDICT_FALSE(++cursor == totalDocIdNum)) {
lastAccessedDocId = NO_MORE_DOCS;
} else {
int iterBlockIndex = cursor >> BLOCK_SIZE_BIT;
int offset = cursor & BLOCK_SIZE_MODULO;
if( iterBlockIndex == compBlockNum ) {
lastAccessedDocId = set->currentNoCompBlock[offset];
} else {
if (PREDICT_TRUE(offset)){
//lastAccessedDocId = iterDecompBlock[offset];
#ifdef PREFIX_SUM
lastAccessedDocId += (iterDecompBlock[offset]);
#else
lastAccessedDocId = iterDecompBlock[offset];
#endif
} else {
// (offset==0) must be in one of the compressed blocks
Source src = set->sequenceOfCompBlocks.get(iterBlockIndex).getSource();
set->codec.Uncompress(src, &iterDecompBlock[0], DEFAULT_BATCH_SIZE);
#ifndef PREFIX_SUM
// postProcessBlock not needed if using integrated delta coding
// postProcessBlock(&iterDecompBlock[0], DEFAULT_BATCH_SIZE);
#endif
// assert(uncompSize == DEFAULT_BATCH_SIZE);
lastAccessedDocId = iterDecompBlock[0];
}
}
}
return lastAccessedDocId;
}
int
session_enqueue_dgram_connection (session_t * s,
session_dgram_hdr_t * hdr,
vlib_buffer_t * b, u8 proto, u8 queue_event)
{
int enqueued = 0, rv, in_order_off;
ASSERT (svm_fifo_max_enqueue_prod (s->rx_fifo)
>= b->current_length + sizeof (*hdr));
svm_fifo_enqueue (s->rx_fifo, sizeof (session_dgram_hdr_t), (u8 *) hdr);
enqueued = svm_fifo_enqueue (s->rx_fifo, b->current_length,
vlib_buffer_get_current (b));
if (PREDICT_FALSE ((b->flags & VLIB_BUFFER_NEXT_PRESENT) && enqueued >= 0))
{
in_order_off = enqueued > b->current_length ? enqueued : 0;
rv = session_enqueue_chain_tail (s, b, in_order_off, 1);
if (rv > 0)
enqueued += rv;
}
if (queue_event)
{
/* Queue RX event on this fifo. Eventually these will need to be flushed
* by calling stream_server_flush_enqueue_events () */
session_worker_t *wrk;
wrk = session_main_get_worker (s->thread_index);
if (!(s->flags & SESSION_F_RX_EVT))
{
s->flags |= SESSION_F_RX_EVT;
vec_add1 (wrk->session_to_enqueue[proto], s->session_index);
}
}
return enqueued;
}
replication_context_t *
replication_prep (vlib_main_t * vm,
vlib_buffer_t * b0,
u32 recycle_node_index,
u32 l2_packet)
{
replication_main_t * rm = &replication_main;
replication_context_t * ctx;
uword cpu_number = vm->cpu_index;
ip4_header_t * ip;
u32 ctx_id;
// Allocate a context, reserve context 0
if (PREDICT_FALSE(rm->contexts[cpu_number] == 0))
pool_get_aligned (rm->contexts[cpu_number], ctx, CLIB_CACHE_LINE_BYTES);
pool_get_aligned (rm->contexts[cpu_number], ctx, CLIB_CACHE_LINE_BYTES);
ctx_id = ctx - rm->contexts[cpu_number];
// Save state from vlib buffer
ctx->saved_clone_count = b0->clone_count;
ctx->saved_free_list_index = b0->free_list_index;
ctx->current_data = b0->current_data;
// Set up vlib buffer hooks
b0->clone_count = ctx_id;
b0->free_list_index = rm->recycle_list_index;
// Save feature state
ctx->recycle_node_index = recycle_node_index;
// Save vnet state
memcpy (ctx->vnet_buffer, vnet_buffer(b0), sizeof(vnet_buffer_opaque_t));
// Save packet contents
ctx->l2_packet = l2_packet;
ip = (ip4_header_t *)vlib_buffer_get_current (b0);
if (l2_packet) {
// Save ethernet header
ctx->l2_header[0] = ((u64 *)ip)[0];
ctx->l2_header[1] = ((u64 *)ip)[1];
ctx->l2_header[2] = ((u64 *)ip)[2];
// set ip to the true ip header
ip = (ip4_header_t *)(((u8 *)ip) + vnet_buffer(b0)->l2.l2_len);
}
// Copy L3 fields.
// We need to save TOS for ip4 and ip6 packets. Fortunately the TOS field is
// in the first two bytes of both the ip4 and ip6 headers.
ctx->ip_tos = *((u16 *)(ip));
// Save the ip4 checksum as well. We just blindly save the corresponding two
// bytes even for ip6 packets.
ctx->ip4_checksum = ip->checksum;
return ctx;
}
void cnat_portmap_dump_v2 (cnat_portmap_v2_t *pm, u16 print_limit)
{
int i;
u32 inuse =0;
ASSERT(pm);
for (i = 0; i < BITS_PER_INST; i++) {
if (PREDICT_FALSE(clib_bitmap_get_no_check (pm->bm, i) == 0)) {
if (PREDICT_TRUE(inuse++ < print_limit))
PLATFORM_DEBUG_PRINT(" %d", bit2port(i));
}
}
if (PREDICT_FALSE(inuse >= print_limit)) {
PLATFORM_DEBUG_PRINT("%d printed, print limit is %d\n",
inuse, print_limit);
}
PLATFORM_DEBUG_PRINT("\n");
}
void vl_msg_api_cleanup_handler (void *the_msg)
{
api_main_t *am = &api_main;
u16 id = ntohs(*((u16 *)the_msg));
if (PREDICT_FALSE(id >= vec_len(am->msg_cleanup_handlers))) {
clib_warning ("_vl_msg_id too large: %d\n", id);
return;
}
if (am->msg_cleanup_handlers[id])
(*am->msg_cleanup_handlers[id])(the_msg);
vl_msg_api_free(the_msg);
}
/*
* vl_msg_api_replay_handler
*/
void vl_msg_api_replay_handler(void *the_msg)
{
api_main_t *am = &api_main;
u16 id = ntohs(*((u16 *)the_msg));
if (PREDICT_FALSE(id >= vec_len(am->msg_handlers))) {
clib_warning ("_vl_msg_id too large: %d\n", id);
return;
}
/* do NOT trace the message... */
if (am->msg_handlers[id])
(*am->msg_handlers[id])(the_msg);
/* do NOT free the message buffer... */
}
//This method will not work after a call to flush()
inline bool CompressedSet::find(unsigned int target) const {
vector<uint32_t,AlignedSTLAllocator<uint32_t, 64>> myDecompBlock(DEFAULT_BATCH_SIZE, 0);
//unsigned int lastId = lastAdded;
if(PREDICT_FALSE(totalDocIdNum==0))
return false;
if (sizeOfCurrentNoCompBlock!=0){
//int lastId = currentNoCompBlock[sizeOfCurrentNoCompBlock-1];
if(sizeOfCurrentNoCompBlock > 0 && target > currentNoCompBlock[sizeOfCurrentNoCompBlock-1])
{
return false;
}
// first search noComp block
if(baseListForOnlyCompBlocks.size()==0 || target>baseListForOnlyCompBlocks[baseListForOnlyCompBlocks.size()-1])
{
int i;
for(i=0;i<sizeOfCurrentNoCompBlock;++i)
{
if(currentNoCompBlock[i] >= target)
break;
}
if(i == sizeOfCurrentNoCompBlock)
return false;
return currentNoCompBlock[i] == target;
}
}
// if we have some CompBlocks
// first find which block to decompress by looking into baseListForOnlyCompBlocks
if(baseListForOnlyCompBlocks.size()>0) {
// baseListForOnlyCompBlocks.size() must then >0
int index = binarySearchInBaseListForBlockThatMayContainTarget(baseListForOnlyCompBlocks, 0, baseListForOnlyCompBlocks.size()-1, target);
if(index<0)
return false;// target is bigger then biggest value
////uncompress block
Source src = sequenceOfCompBlocks.get(index).getSource();
size_t uncompSize = codec.Uncompress(src, &myDecompBlock[0] ,DEFAULT_BATCH_SIZE);
return codec.findInArray(&myDecompBlock[0], uncompSize,target);
}
return false;
}
void
vl_msg_api_send_shmem (svm_queue_t * q, u8 * elem)
{
api_main_t *am = &api_main;
uword *trace = (uword *) elem;
if (am->tx_trace && am->tx_trace->enabled)
vl_msg_api_trace (am, am->tx_trace, (void *) trace[0]);
/*
* Announce a probable binary API client bug:
* some client's input queue is stuffed.
* The situation may be recoverable, or not.
*/
if (PREDICT_FALSE
(am->vl_clients /* vpp side */ && (q->cursize == q->maxsize)))
clib_warning ("WARNING: client input queue at %llx is stuffed...", q);
(void) svm_queue_add (q, elem, 0 /* nowait */ );
}
static inline u32
is_pcp_pkt(spp_ctx_t *ctx, u32 addr, u16 port)
{
cnat_vrfmap_t *my_vrfmap = NULL;
u16 my_vrfmap_index;
my_vrfmap_index = vrf_map_array[ctx->ru.rx.uidb_index];
if (PREDICT_TRUE(my_vrfmap_index != VRF_MAP_ENTRY_EMPTY)) {
my_vrfmap = cnat_map_by_vrf + my_vrfmap_index;
if (PREDICT_FALSE( port == my_vrfmap->pcp_server_port)) {
if(PREDICT_TRUE(addr == my_vrfmap->pcp_server_addr)) {
return CNAT_SUCCESS;
}
}
}
return CNAT_NO_CONFIG;
}
/**
* Notify session peer that new data has been enqueued.
*
* @param s Stream session for which the event is to be generated.
* @param lock Flag to indicate if call should lock message queue.
*
* @return 0 on success or negative number if failed to send notification.
*/
static inline int
session_enqueue_notify_inline (session_t * s)
{
app_worker_t *app_wrk;
u32 session_index;
u8 n_subscribers;
session_index = s->session_index;
n_subscribers = svm_fifo_n_subscribers (s->rx_fifo);
app_wrk = app_worker_get_if_valid (s->app_wrk_index);
if (PREDICT_FALSE (!app_wrk))
{
SESSION_DBG ("invalid s->app_index = %d", s->app_wrk_index);
return 0;
}
/* *INDENT-OFF* */
SESSION_EVT_DBG(SESSION_EVT_ENQ, s, ({
ed->data[0] = SESSION_IO_EVT_RX;
ed->data[1] = svm_fifo_max_dequeue_prod (s->rx_fifo);
}));
session_t *
session_alloc (u32 thread_index)
{
session_worker_t *wrk = &session_main.wrk[thread_index];
session_t *s;
u8 will_expand = 0;
pool_get_aligned_will_expand (wrk->sessions, will_expand,
CLIB_CACHE_LINE_BYTES);
/* If we have peekers, let them finish */
if (PREDICT_FALSE (will_expand && vlib_num_workers ()))
{
clib_rwlock_writer_lock (&wrk->peekers_rw_locks);
pool_get_aligned (wrk->sessions, s, CLIB_CACHE_LINE_BYTES);
clib_rwlock_writer_unlock (&wrk->peekers_rw_locks);
}
else
{
pool_get_aligned (wrk->sessions, s, CLIB_CACHE_LINE_BYTES);
}
clib_memset (s, 0, sizeof (*s));
s->session_index = s - wrk->sessions;
s->thread_index = thread_index;
return s;
}
/*
* cnat_port_alloc_v2
* public ipv4 address/port allocator for dynamic ports
*
* 200K users / 20M translations means vec_len(cnat_portmap) will be
* around 300.
*
*/
cnat_errno_t
cnat_dynamic_port_alloc_v2 (
cnat_portmap_v2_t *pm,
port_alloc_t atype,
port_pair_t pair_type,
u32 *index,
u32 *o_ipv4_address,
u16 *o_port,
u16 static_port_range
#ifndef NO_BULK_LOGGING
, bulk_alloc_size_t bulk_size,
int *nfv9_log_req
#endif
, u16 ip_n_to_1,
u32 *rseed_ip
)
{
int i;
cnat_errno_t my_err = CNAT_NO_POOL_ANY;
cnat_portmap_v2_t *my_pm = 0;
u16 start_bit;
u16 new_port;
uword bit_test_result;
uword max_trys_to_find_port;
ASSERT(index);
ASSERT(o_ipv4_address);
ASSERT(o_port);
my_pm = cnat_dynamic_addr_alloc_from_pm(pm, atype, index, &my_err, ip_n_to_1,
rseed_ip);
if (PREDICT_FALSE(my_pm == NULL)) {
return (my_err);
}
if(PREDICT_FALSE(my_pm->dyn_full == 1)) {
if (atype == PORT_ALLOC_DIRECTED) {
return (CNAT_NOT_FOUND_DIRECT);
} else {
return (CNAT_NOT_FOUND_ANY);
}
}
#if DEBUG > 1
PLATFORM_DEBUG_PRINT("ALLOC_PORT_V2: My_Instance_Number %d: IP addr 0x%x, Inuse %d\n",
my_instance_number, my_pm->ipv4_address, my_pm->inuse);
#endif
rseed_port = randq1(rseed_port);
/*
* Exclude the static port range for allocating dynamic ports
*/
start_bit = (rseed_port) % (BITS_PER_INST - static_port_range);
start_bit = start_bit + static_port_range;
#ifndef NO_BULK_LOGGING
*nfv9_log_req = BULK_ALLOC_NOT_ATTEMPTED;
if(BULK_ALLOC_SIZE_NONE != bulk_size)
{
/* We need the start port of the range to be alined on integer multiple
* of bulk_size */
max_trys_to_find_port = BITS_PER_INST/bulk_size;
start_bit= ((start_bit + bulk_size -1)/bulk_size) * bulk_size;
}
else
#endif /* #ifndef NO_BULK_LOGGING */
max_trys_to_find_port = BITS_PER_INST;
/* Allocate a random port / port-pair */
for (i = 0; i < max_trys_to_find_port; i++) {
/* start_bit is only a u16.. so it can rollover and become zero */
if (PREDICT_FALSE((start_bit >= BITS_PER_INST) ||
(start_bit < static_port_range))) {
start_bit = static_port_range;
#ifndef NO_BULK_LOGGING
if(BULK_ALLOC_SIZE_NONE != bulk_size) {
start_bit= ((start_bit + bulk_size -1)/bulk_size) * bulk_size;
}
#endif /* #ifndef NO_BULK_LOGGING */
}
/* Scan forward from random position */
#ifndef NO_BULK_LOGGING
if(BULK_ALLOC_SIZE_NONE != bulk_size) {
bit_test_result = cgn_clib_bitmap_check_if_all(my_pm->bm,
start_bit, bulk_size);
}
else
#endif /* #ifndef NO_BULK_LOGGING */
bit_test_result = clib_bitmap_get_no_check(my_pm->bm, start_bit);
if (PREDICT_TRUE(bit_test_result)) {
new_port = bit2port(start_bit);
#ifndef NO_BULK_LOGGING
if(BULK_ALLOC_SIZE_NONE != bulk_size)
*nfv9_log_req = new_port;
#endif
if ((pair_type == PORT_S_ODD) &&
(!(new_port & 0x1))) {
#ifndef NO_BULK_LOGGING
if(BULK_ALLOC_SIZE_NONE != bulk_size) {
start_bit++; /* Just use the next one in the bulk range */
new_port++;
goto found2;
}
#endif /* #ifndef NO_BULK_LOGGING */
goto notfound;
} else if ((pair_type == PORT_S_EVEN) &&
(new_port & 0x1)) {
goto notfound;
}
/* OK we got one or two suitable ports */
goto found2;
}
notfound:
#ifndef NO_BULK_LOGGING
if(BULK_ALLOC_SIZE_NONE != bulk_size)
start_bit += bulk_size;
else
#endif /* #ifndef NO_BULK_LOGGING */
start_bit++;
} /* end of for loop */
/* Completely out of ports */
/* Port allocation failure */
/* set dyn_full flag. This would be used to verify
* for further dyn session before searching for port
*/
if (atype == PORT_ALLOC_DIRECTED) {
my_pm->dyn_full = 1;
return (CNAT_NOT_FOUND_DIRECT);
} else {
my_pm->dyn_full = 1;
return (CNAT_NOT_FOUND_ANY);
}
found2:
/* Accounting */
cgn_clib_bitmap_clear_no_check (my_pm->bm, start_bit);
(my_pm->inuse)++;
*index = my_pm - pm;
*o_ipv4_address = my_pm->ipv4_address;
*o_port = new_port;
return (CNAT_SUCCESS);
}
always_inline uword
geneve_encap_inline (vlib_main_t * vm,
vlib_node_runtime_t * node,
vlib_frame_t * from_frame, u32 is_ip4)
{
u32 n_left_from, next_index, *from, *to_next;
geneve_main_t *vxm = &geneve_main;
vnet_main_t *vnm = vxm->vnet_main;
vnet_interface_main_t *im = &vnm->interface_main;
u32 pkts_encapsulated = 0;
u16 old_l0 = 0, old_l1 = 0;
u32 thread_index = vm->thread_index;
u32 stats_sw_if_index, stats_n_packets, stats_n_bytes;
u32 sw_if_index0 = ~0, sw_if_index1 = ~0;
u32 next0 = 0, next1 = 0;
vnet_hw_interface_t *hi0, *hi1;
geneve_tunnel_t *t0 = NULL, *t1 = NULL;
from = vlib_frame_vector_args (from_frame);
n_left_from = from_frame->n_vectors;
next_index = node->cached_next_index;
stats_sw_if_index = node->runtime_data[0];
stats_n_packets = stats_n_bytes = 0;
while (n_left_from > 0)
{
u32 n_left_to_next;
vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
while (n_left_from >= 4 && n_left_to_next >= 2)
{
u32 bi0, bi1;
vlib_buffer_t *b0, *b1;
u32 flow_hash0, flow_hash1;
u32 len0, len1;
ip4_header_t *ip4_0, *ip4_1;
ip6_header_t *ip6_0, *ip6_1;
udp_header_t *udp0, *udp1;
u64 *copy_src0, *copy_dst0;
u64 *copy_src1, *copy_dst1;
u32 *copy_src_last0, *copy_dst_last0;
u32 *copy_src_last1, *copy_dst_last1;
u16 new_l0, new_l1;
ip_csum_t sum0, sum1;
/* 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, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
CLIB_PREFETCH (p3->data, 2 * CLIB_CACHE_LINE_BYTES, LOAD);
}
bi0 = from[0];
bi1 = from[1];
to_next[0] = bi0;
to_next[1] = bi1;
from += 2;
to_next += 2;
n_left_to_next -= 2;
n_left_from -= 2;
b0 = vlib_get_buffer (vm, bi0);
b1 = vlib_get_buffer (vm, bi1);
flow_hash0 = vnet_l2_compute_flow_hash (b0);
flow_hash1 = vnet_l2_compute_flow_hash (b1);
/* Get next node index and adj index from tunnel next_dpo */
if (sw_if_index0 != vnet_buffer (b0)->sw_if_index[VLIB_TX])
{
sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_TX];
hi0 = vnet_get_sup_hw_interface (vnm, sw_if_index0);
t0 = &vxm->tunnels[hi0->dev_instance];
/* Note: change to always set next0 if it may be set to drop */
next0 = t0->next_dpo.dpoi_next_node;
}
ASSERT (t0 != NULL);
vnet_buffer (b0)->ip.adj_index[VLIB_TX] = t0->next_dpo.dpoi_index;
/* Get next node index and adj index from tunnel next_dpo */
if (sw_if_index1 != vnet_buffer (b1)->sw_if_index[VLIB_TX])
{
sw_if_index1 = vnet_buffer (b1)->sw_if_index[VLIB_TX];
hi1 = vnet_get_sup_hw_interface (vnm, sw_if_index1);
t1 = &vxm->tunnels[hi1->dev_instance];
/* Note: change to always set next1 if it may be set to drop */
next1 = t1->next_dpo.dpoi_next_node;
}
ASSERT (t1 != NULL);
vnet_buffer (b1)->ip.adj_index[VLIB_TX] = t1->next_dpo.dpoi_index;
/* Apply the rewrite string. $$$$ vnet_rewrite? */
vlib_buffer_advance (b0, -(word) _vec_len (t0->rewrite));
vlib_buffer_advance (b1, -(word) _vec_len (t1->rewrite));
if (is_ip4)
{
u8 ip4_geneve_base_header_len =
sizeof (ip4_header_t) + sizeof (udp_header_t) +
GENEVE_BASE_HEADER_LENGTH;
u8 ip4_geneve_header_total_len0 = ip4_geneve_base_header_len;
u8 ip4_geneve_header_total_len1 = ip4_geneve_base_header_len;
#if SUPPORT_OPTIONS_HEADER==1
ip4_geneve_header_total_len0 += t0->options_len;
ip4_geneve_header_total_len1 += t1->options_len;
#endif
ASSERT (vec_len (t0->rewrite) == ip4_geneve_header_total_len0);
ASSERT (vec_len (t1->rewrite) == ip4_geneve_header_total_len1);
ip4_0 = vlib_buffer_get_current (b0);
ip4_1 = vlib_buffer_get_current (b1);
/* Copy the fixed header */
copy_dst0 = (u64 *) ip4_0;
copy_src0 = (u64 *) t0->rewrite;
copy_dst1 = (u64 *) ip4_1;
copy_src1 = (u64 *) t1->rewrite;
/* Copy first 32 octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
foreach_fixed_header4_offset;
#undef _
#define _(offs) copy_dst1[offs] = copy_src1[offs];
foreach_fixed_header4_offset;
#undef _
/* Last 4 octets. Hopefully gcc will be our friend */
copy_dst_last0 = (u32 *) (©_dst0[4]);
copy_src_last0 = (u32 *) (©_src0[4]);
copy_dst_last0[0] = copy_src_last0[0];
copy_dst_last1 = (u32 *) (©_dst1[4]);
copy_src_last1 = (u32 *) (©_src1[4]);
copy_dst_last1[0] = copy_src_last1[0];
/* Fix the IP4 checksum and length */
sum0 = ip4_0->checksum;
new_l0 = /* old_l0 always 0, see the rewrite setup */
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0));
sum0 = ip_csum_update (sum0, old_l0, new_l0, ip4_header_t,
length /* changed member */ );
ip4_0->checksum = ip_csum_fold (sum0);
ip4_0->length = new_l0;
sum1 = ip4_1->checksum;
new_l1 = /* old_l1 always 0, see the rewrite setup */
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1));
sum1 = ip_csum_update (sum1, old_l1, new_l1, ip4_header_t,
length /* changed member */ );
ip4_1->checksum = ip_csum_fold (sum1);
ip4_1->length = new_l1;
/* Fix UDP length and set source port */
udp0 = (udp_header_t *) (ip4_0 + 1);
new_l0 =
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0) -
sizeof (*ip4_0));
udp0->length = new_l0;
udp0->src_port = flow_hash0;
udp1 = (udp_header_t *) (ip4_1 + 1);
new_l1 =
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1) -
sizeof (*ip4_1));
udp1->length = new_l1;
udp1->src_port = flow_hash1;
}
else /* ipv6 */
{
int bogus = 0;
u8 ip6_geneve_base_header_len =
sizeof (ip6_header_t) + sizeof (udp_header_t) +
GENEVE_BASE_HEADER_LENGTH;
u8 ip6_geneve_header_total_len0 = ip6_geneve_base_header_len;
u8 ip6_geneve_header_total_len1 = ip6_geneve_base_header_len;
#if SUPPORT_OPTIONS_HEADER==1
ip6_geneve_header_total_len0 += t0->options_len;
ip6_geneve_header_total_len1 += t1->options_len;
#endif
ASSERT (vec_len (t0->rewrite) == ip6_geneve_header_total_len0);
ASSERT (vec_len (t1->rewrite) == ip6_geneve_header_total_len1);
ip6_0 = vlib_buffer_get_current (b0);
ip6_1 = vlib_buffer_get_current (b1);
/* Copy the fixed header */
copy_dst0 = (u64 *) ip6_0;
copy_src0 = (u64 *) t0->rewrite;
copy_dst1 = (u64 *) ip6_1;
copy_src1 = (u64 *) t1->rewrite;
/* Copy first 56 (ip6) octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
foreach_fixed_header6_offset;
#undef _
#define _(offs) copy_dst1[offs] = copy_src1[offs];
foreach_fixed_header6_offset;
#undef _
/* Fix IP6 payload length */
new_l0 =
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0)
- sizeof (*ip6_0));
ip6_0->payload_length = new_l0;
new_l1 =
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1)
- sizeof (*ip6_1));
ip6_1->payload_length = new_l1;
/* Fix UDP length and set source port */
udp0 = (udp_header_t *) (ip6_0 + 1);
udp0->length = new_l0;
udp0->src_port = flow_hash0;
udp1 = (udp_header_t *) (ip6_1 + 1);
udp1->length = new_l1;
udp1->src_port = flow_hash1;
/* IPv6 UDP checksum is mandatory */
udp0->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b0,
ip6_0,
&bogus);
ASSERT (bogus == 0);
if (udp0->checksum == 0)
udp0->checksum = 0xffff;
udp1->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b1,
ip6_1,
&bogus);
ASSERT (bogus == 0);
if (udp1->checksum == 0)
udp1->checksum = 0xffff;
}
pkts_encapsulated += 2;
len0 = vlib_buffer_length_in_chain (vm, b0);
len1 = vlib_buffer_length_in_chain (vm, b1);
stats_n_packets += 2;
stats_n_bytes += len0 + len1;
/* Batch stats increment on the same geneve tunnel so counter is not
incremented per packet. Note stats are still incremented for deleted
and admin-down tunnel where packets are dropped. It is not worthwhile
to check for this rare case and affect normal path performance. */
if (PREDICT_FALSE ((sw_if_index0 != stats_sw_if_index) ||
(sw_if_index1 != stats_sw_if_index)))
{
stats_n_packets -= 2;
stats_n_bytes -= len0 + len1;
if (sw_if_index0 == sw_if_index1)
{
if (stats_n_packets)
vlib_increment_combined_counter
(im->combined_sw_if_counters +
VNET_INTERFACE_COUNTER_TX, thread_index,
stats_sw_if_index, stats_n_packets, stats_n_bytes);
stats_sw_if_index = sw_if_index0;
stats_n_packets = 2;
stats_n_bytes = len0 + len1;
}
else
{
vlib_increment_combined_counter
(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
thread_index, sw_if_index0, 1, len0);
vlib_increment_combined_counter
(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
thread_index, sw_if_index1, 1, len1);
}
}
if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
{
geneve_encap_trace_t *tr =
vlib_add_trace (vm, node, b0, sizeof (*tr));
tr->tunnel_index = t0 - vxm->tunnels;
tr->vni = t0->vni;
}
if (PREDICT_FALSE (b1->flags & VLIB_BUFFER_IS_TRACED))
{
geneve_encap_trace_t *tr =
vlib_add_trace (vm, node, b1, sizeof (*tr));
tr->tunnel_index = t1 - vxm->tunnels;
tr->vni = t1->vni;
}
vlib_validate_buffer_enqueue_x2 (vm, node, next_index,
to_next, n_left_to_next,
bi0, bi1, next0, next1);
}
while (n_left_from > 0 && n_left_to_next > 0)
{
u32 bi0;
vlib_buffer_t *b0;
u32 flow_hash0;
u32 len0;
ip4_header_t *ip4_0;
ip6_header_t *ip6_0;
udp_header_t *udp0;
u64 *copy_src0, *copy_dst0;
u32 *copy_src_last0, *copy_dst_last0;
u16 new_l0;
ip_csum_t sum0;
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);
flow_hash0 = vnet_l2_compute_flow_hash (b0);
/* Get next node index and adj index from tunnel next_dpo */
if (sw_if_index0 != vnet_buffer (b0)->sw_if_index[VLIB_TX])
{
sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_TX];
hi0 = vnet_get_sup_hw_interface (vnm, sw_if_index0);
t0 = &vxm->tunnels[hi0->dev_instance];
/* Note: change to always set next0 if it may be set to drop */
next0 = t0->next_dpo.dpoi_next_node;
}
vnet_buffer (b0)->ip.adj_index[VLIB_TX] = t0->next_dpo.dpoi_index;
/* Apply the rewrite string. $$$$ vnet_rewrite? */
vlib_buffer_advance (b0, -(word) _vec_len (t0->rewrite));
if (is_ip4)
{
u8 ip4_geneve_base_header_len =
sizeof (ip4_header_t) + sizeof (udp_header_t) +
GENEVE_BASE_HEADER_LENGTH;
u8 ip4_geneve_header_total_len0 = ip4_geneve_base_header_len;
#if SUPPORT_OPTIONS_HEADER==1
ip4_geneve_header_total_len0 += t0->options_len;
#endif
ASSERT (vec_len (t0->rewrite) == ip4_geneve_header_total_len0);
ip4_0 = vlib_buffer_get_current (b0);
/* Copy the fixed header */
copy_dst0 = (u64 *) ip4_0;
copy_src0 = (u64 *) t0->rewrite;
/* Copy first 32 octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
foreach_fixed_header4_offset;
#undef _
/* Last 4 octets. Hopefully gcc will be our friend */
copy_dst_last0 = (u32 *) (©_dst0[4]);
copy_src_last0 = (u32 *) (©_src0[4]);
copy_dst_last0[0] = copy_src_last0[0];
/* Fix the IP4 checksum and length */
sum0 = ip4_0->checksum;
new_l0 = /* old_l0 always 0, see the rewrite setup */
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0));
sum0 = ip_csum_update (sum0, old_l0, new_l0, ip4_header_t,
length /* changed member */ );
ip4_0->checksum = ip_csum_fold (sum0);
ip4_0->length = new_l0;
/* Fix UDP length and set source port */
udp0 = (udp_header_t *) (ip4_0 + 1);
new_l0 =
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0) -
sizeof (*ip4_0));
udp0->length = new_l0;
udp0->src_port = flow_hash0;
}
else /* ip6 path */
{
int bogus = 0;
u8 ip6_geneve_base_header_len =
sizeof (ip6_header_t) + sizeof (udp_header_t) +
GENEVE_BASE_HEADER_LENGTH;
u8 ip6_geneve_header_total_len0 = ip6_geneve_base_header_len;
#if SUPPORT_OPTIONS_HEADER==1
ip6_geneve_header_total_len0 += t0->options_len;
#endif
ASSERT (vec_len (t0->rewrite) == ip6_geneve_header_total_len0);
ip6_0 = vlib_buffer_get_current (b0);
/* Copy the fixed header */
copy_dst0 = (u64 *) ip6_0;
copy_src0 = (u64 *) t0->rewrite;
/* Copy first 56 (ip6) octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
foreach_fixed_header6_offset;
#undef _
/* Fix IP6 payload length */
new_l0 =
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0)
- sizeof (*ip6_0));
ip6_0->payload_length = new_l0;
/* Fix UDP length and set source port */
udp0 = (udp_header_t *) (ip6_0 + 1);
udp0->length = new_l0;
udp0->src_port = flow_hash0;
/* IPv6 UDP checksum is mandatory */
udp0->checksum = ip6_tcp_udp_icmp_compute_checksum (vm, b0,
ip6_0,
&bogus);
ASSERT (bogus == 0);
if (udp0->checksum == 0)
udp0->checksum = 0xffff;
}
pkts_encapsulated++;
len0 = vlib_buffer_length_in_chain (vm, b0);
stats_n_packets += 1;
stats_n_bytes += len0;
/* Batch stats increment on the same geneve tunnel so counter is not
incremented per packet. Note stats are still incremented for deleted
and admin-down tunnel where packets are dropped. It is not worthwhile
to check for this rare case and affect normal path performance. */
if (PREDICT_FALSE (sw_if_index0 != stats_sw_if_index))
{
stats_n_packets -= 1;
stats_n_bytes -= len0;
if (stats_n_packets)
vlib_increment_combined_counter
(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
thread_index, stats_sw_if_index,
stats_n_packets, stats_n_bytes);
stats_n_packets = 1;
stats_n_bytes = len0;
stats_sw_if_index = sw_if_index0;
}
if (PREDICT_FALSE (b0->flags & VLIB_BUFFER_IS_TRACED))
{
geneve_encap_trace_t *tr =
vlib_add_trace (vm, node, b0, sizeof (*tr));
tr->tunnel_index = t0 - vxm->tunnels;
tr->vni = t0->vni;
}
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);
}
/* Do we still need this now that tunnel tx stats is kept? */
vlib_node_increment_counter (vm, node->node_index,
GENEVE_ENCAP_ERROR_ENCAPSULATED,
pkts_encapsulated);
/* Increment any remaining batch stats */
if (stats_n_packets)
{
vlib_increment_combined_counter
(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
thread_index, stats_sw_if_index, stats_n_packets, stats_n_bytes);
node->runtime_data[0] = stats_sw_if_index;
}
return from_frame->n_vectors;
}
always_inline uword
gtpu_encap_inline (vlib_main_t * vm,
vlib_node_runtime_t * node,
vlib_frame_t * from_frame,
u32 is_ip4)
{
u32 n_left_from, next_index, * from, * to_next;
gtpu_main_t * gtm = >pu_main;
vnet_main_t * vnm = gtm->vnet_main;
vnet_interface_main_t * im = &vnm->interface_main;
u32 pkts_encapsulated = 0;
u16 old_l0 = 0, old_l1 = 0, old_l2 = 0, old_l3 = 0;
u32 thread_index = vlib_get_thread_index();
u32 stats_sw_if_index, stats_n_packets, stats_n_bytes;
u32 sw_if_index0 = 0, sw_if_index1 = 0, sw_if_index2 = 0, sw_if_index3 = 0;
u32 next0 = 0, next1 = 0, next2 = 0, next3 = 0;
vnet_hw_interface_t * hi0, * hi1, * hi2, * hi3;
gtpu_tunnel_t * t0 = NULL, * t1 = NULL, * t2 = NULL, * t3 = NULL;
from = vlib_frame_vector_args (from_frame);
n_left_from = from_frame->n_vectors;
next_index = node->cached_next_index;
stats_sw_if_index = node->runtime_data[0];
stats_n_packets = stats_n_bytes = 0;
while (n_left_from > 0)
{
u32 n_left_to_next;
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, bi1, bi2, bi3;
vlib_buffer_t * b0, * b1, * b2, * b3;
u32 flow_hash0, flow_hash1, flow_hash2, flow_hash3;
u32 len0, len1, len2, len3;
ip4_header_t * ip4_0, * ip4_1, * ip4_2, * ip4_3;
ip6_header_t * ip6_0, * ip6_1, * ip6_2, * ip6_3;
udp_header_t * udp0, * udp1, * udp2, * udp3;
gtpu_header_t * gtpu0, * gtpu1, * gtpu2, * gtpu3;
u64 * copy_src0, * copy_dst0;
u64 * copy_src1, * copy_dst1;
u64 * copy_src2, * copy_dst2;
u64 * copy_src3, * copy_dst3;
u32 * copy_src_last0, * copy_dst_last0;
u32 * copy_src_last1, * copy_dst_last1;
u32 * copy_src_last2, * copy_dst_last2;
u32 * copy_src_last3, * copy_dst_last3;
u16 new_l0, new_l1, new_l2, new_l3;
ip_csum_t sum0, sum1, sum2, sum3;
/* Prefetch next iteration. */
{
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, 2*CLIB_CACHE_LINE_BYTES, LOAD);
CLIB_PREFETCH (p5->data, 2*CLIB_CACHE_LINE_BYTES, LOAD);
CLIB_PREFETCH (p6->data, 2*CLIB_CACHE_LINE_BYTES, LOAD);
CLIB_PREFETCH (p7->data, 2*CLIB_CACHE_LINE_BYTES, LOAD);
}
bi0 = from[0];
bi1 = from[1];
bi2 = from[2];
bi3 = from[3];
to_next[0] = bi0;
to_next[1] = bi1;
to_next[2] = bi2;
to_next[3] = bi3;
from += 4;
to_next += 4;
n_left_to_next -= 4;
n_left_from -= 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);
flow_hash0 = vnet_l2_compute_flow_hash (b0);
flow_hash1 = vnet_l2_compute_flow_hash (b1);
flow_hash2 = vnet_l2_compute_flow_hash (b2);
flow_hash3 = vnet_l2_compute_flow_hash (b3);
/* Get next node index and adj index from tunnel next_dpo */
sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX];
sw_if_index1 = vnet_buffer(b1)->sw_if_index[VLIB_TX];
sw_if_index2 = vnet_buffer(b2)->sw_if_index[VLIB_TX];
sw_if_index3 = vnet_buffer(b3)->sw_if_index[VLIB_TX];
hi0 = vnet_get_sup_hw_interface (vnm, sw_if_index0);
hi1 = vnet_get_sup_hw_interface (vnm, sw_if_index1);
hi2 = vnet_get_sup_hw_interface (vnm, sw_if_index2);
hi3 = vnet_get_sup_hw_interface (vnm, sw_if_index3);
t0 = >m->tunnels[hi0->dev_instance];
t1 = >m->tunnels[hi1->dev_instance];
t2 = >m->tunnels[hi2->dev_instance];
t3 = >m->tunnels[hi3->dev_instance];
/* Note: change to always set next0 if it may be set to drop */
next0 = t0->next_dpo.dpoi_next_node;
vnet_buffer(b0)->ip.adj_index[VLIB_TX] = t0->next_dpo.dpoi_index;
next1 = t1->next_dpo.dpoi_next_node;
vnet_buffer(b1)->ip.adj_index[VLIB_TX] = t1->next_dpo.dpoi_index;
next2 = t2->next_dpo.dpoi_next_node;
vnet_buffer(b2)->ip.adj_index[VLIB_TX] = t2->next_dpo.dpoi_index;
next3 = t3->next_dpo.dpoi_next_node;
vnet_buffer(b3)->ip.adj_index[VLIB_TX] = t3->next_dpo.dpoi_index;
/* Apply the rewrite string. $$$$ vnet_rewrite? */
vlib_buffer_advance (b0, -(word)_vec_len(t0->rewrite));
vlib_buffer_advance (b1, -(word)_vec_len(t1->rewrite));
vlib_buffer_advance (b2, -(word)_vec_len(t2->rewrite));
vlib_buffer_advance (b3, -(word)_vec_len(t3->rewrite));
if (is_ip4)
{
ip4_0 = vlib_buffer_get_current(b0);
ip4_1 = vlib_buffer_get_current(b1);
ip4_2 = vlib_buffer_get_current(b2);
ip4_3 = vlib_buffer_get_current(b3);
/* Copy the fixed header */
copy_dst0 = (u64 *) ip4_0;
copy_src0 = (u64 *) t0->rewrite;
copy_dst1 = (u64 *) ip4_1;
copy_src1 = (u64 *) t1->rewrite;
copy_dst2 = (u64 *) ip4_2;
copy_src2 = (u64 *) t2->rewrite;
copy_dst3 = (u64 *) ip4_3;
copy_src3 = (u64 *) t3->rewrite;
/* Copy first 32 octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
foreach_fixed_header4_offset;
#undef _
#define _(offs) copy_dst1[offs] = copy_src1[offs];
foreach_fixed_header4_offset;
#undef _
#define _(offs) copy_dst2[offs] = copy_src2[offs];
foreach_fixed_header4_offset;
#undef _
#define _(offs) copy_dst3[offs] = copy_src3[offs];
foreach_fixed_header4_offset;
#undef _
/* Last 4 octets. Hopefully gcc will be our friend */
copy_dst_last0 = (u32 *)(©_dst0[4]);
copy_src_last0 = (u32 *)(©_src0[4]);
copy_dst_last0[0] = copy_src_last0[0];
copy_dst_last1 = (u32 *)(©_dst1[4]);
copy_src_last1 = (u32 *)(©_src1[4]);
copy_dst_last1[0] = copy_src_last1[0];
copy_dst_last2 = (u32 *)(©_dst2[4]);
copy_src_last2 = (u32 *)(©_src2[4]);
copy_dst_last2[0] = copy_src_last2[0];
copy_dst_last3 = (u32 *)(©_dst3[4]);
copy_src_last3 = (u32 *)(©_src3[4]);
copy_dst_last3[0] = copy_src_last3[0];
/* Fix the IP4 checksum and length */
sum0 = ip4_0->checksum;
new_l0 = /* old_l0 always 0, see the rewrite setup */
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0));
sum0 = ip_csum_update (sum0, old_l0, new_l0, ip4_header_t,
length /* changed member */);
ip4_0->checksum = ip_csum_fold (sum0);
ip4_0->length = new_l0;
sum1 = ip4_1->checksum;
new_l1 = /* old_l1 always 0, see the rewrite setup */
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1));
sum1 = ip_csum_update (sum1, old_l1, new_l1, ip4_header_t,
length /* changed member */);
ip4_1->checksum = ip_csum_fold (sum1);
ip4_1->length = new_l1;
sum2 = ip4_2->checksum;
new_l2 = /* old_l0 always 0, see the rewrite setup */
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b2));
sum2 = ip_csum_update (sum2, old_l2, new_l2, ip4_header_t,
length /* changed member */);
ip4_2->checksum = ip_csum_fold (sum2);
ip4_2->length = new_l2;
sum3 = ip4_3->checksum;
new_l3 = /* old_l1 always 0, see the rewrite setup */
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b3));
sum3 = ip_csum_update (sum3, old_l3, new_l3, ip4_header_t,
length /* changed member */);
ip4_3->checksum = ip_csum_fold (sum3);
ip4_3->length = new_l3;
/* Fix UDP length and set source port */
udp0 = (udp_header_t *)(ip4_0+1);
new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
- sizeof (*ip4_0));
udp0->length = new_l0;
udp0->src_port = flow_hash0;
udp1 = (udp_header_t *)(ip4_1+1);
new_l1 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b1)
- sizeof (*ip4_1));
udp1->length = new_l1;
udp1->src_port = flow_hash1;
udp2 = (udp_header_t *)(ip4_2+1);
new_l2 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b2)
- sizeof (*ip4_2));
udp2->length = new_l2;
udp2->src_port = flow_hash2;
udp3 = (udp_header_t *)(ip4_3+1);
new_l3 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b3)
- sizeof (*ip4_3));
udp3->length = new_l3;
udp3->src_port = flow_hash3;
/* Fix GTPU length */
gtpu0 = (gtpu_header_t *)(udp0+1);
new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
- sizeof (*ip4_0) - sizeof(*udp0)
- GTPU_V1_HDR_LEN);
gtpu0->length = new_l0;
gtpu1 = (gtpu_header_t *)(udp1+1);
new_l1 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b1)
- sizeof (*ip4_1) - sizeof(*udp1)
- GTPU_V1_HDR_LEN);
gtpu1->length = new_l1;
gtpu2 = (gtpu_header_t *)(udp2+1);
new_l2 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b2)
- sizeof (*ip4_2) - sizeof(*udp2)
- GTPU_V1_HDR_LEN);
gtpu2->length = new_l2;
gtpu3 = (gtpu_header_t *)(udp3+1);
new_l3 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b3)
- sizeof (*ip4_3) - sizeof(*udp3)
- GTPU_V1_HDR_LEN);
gtpu3->length = new_l3;
}
else /* ipv6 */
{
int bogus = 0;
ip6_0 = vlib_buffer_get_current(b0);
ip6_1 = vlib_buffer_get_current(b1);
ip6_2 = vlib_buffer_get_current(b2);
ip6_3 = vlib_buffer_get_current(b3);
/* Copy the fixed header */
copy_dst0 = (u64 *) ip6_0;
copy_src0 = (u64 *) t0->rewrite;
copy_dst1 = (u64 *) ip6_1;
copy_src1 = (u64 *) t1->rewrite;
copy_dst2 = (u64 *) ip6_2;
copy_src2 = (u64 *) t2->rewrite;
copy_dst3 = (u64 *) ip6_3;
copy_src3 = (u64 *) t3->rewrite;
/* Copy first 56 (ip6) octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
foreach_fixed_header6_offset;
#undef _
#define _(offs) copy_dst1[offs] = copy_src1[offs];
foreach_fixed_header6_offset;
#undef _
#define _(offs) copy_dst2[offs] = copy_src2[offs];
foreach_fixed_header6_offset;
#undef _
#define _(offs) copy_dst3[offs] = copy_src3[offs];
foreach_fixed_header6_offset;
#undef _
/* Fix IP6 payload length */
new_l0 =
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0)
- sizeof(*ip6_0));
ip6_0->payload_length = new_l0;
new_l1 =
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b1)
- sizeof(*ip6_1));
ip6_1->payload_length = new_l1;
new_l2 =
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b2)
- sizeof(*ip6_2));
ip6_2->payload_length = new_l2;
new_l3 =
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b3)
- sizeof(*ip6_3));
ip6_3->payload_length = new_l3;
/* Fix UDP length and set source port */
udp0 = (udp_header_t *)(ip6_0+1);
udp0->length = new_l0;
udp0->src_port = flow_hash0;
udp1 = (udp_header_t *)(ip6_1+1);
udp1->length = new_l1;
udp1->src_port = flow_hash1;
udp2 = (udp_header_t *)(ip6_2+1);
udp2->length = new_l2;
udp2->src_port = flow_hash2;
udp3 = (udp_header_t *)(ip6_3+1);
udp3->length = new_l3;
udp3->src_port = flow_hash3;
/* Fix GTPU length */
gtpu0 = (gtpu_header_t *)(udp0+1);
new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
- sizeof (*ip6_0) - sizeof(*udp0)
- GTPU_V1_HDR_LEN);
gtpu0->length = new_l0;
gtpu1 = (gtpu_header_t *)(udp1+1);
new_l1 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b1)
- sizeof (*ip6_1) - sizeof(*udp1)
- GTPU_V1_HDR_LEN);
gtpu1->length = new_l1;
gtpu2 = (gtpu_header_t *)(udp2+1);
new_l2 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b2)
- sizeof (*ip6_2) - sizeof(*udp2)
- GTPU_V1_HDR_LEN);
gtpu2->length = new_l2;
gtpu3 = (gtpu_header_t *)(udp3+1);
new_l3 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b3)
- sizeof (*ip6_3) - sizeof(*udp3)
- GTPU_V1_HDR_LEN);
gtpu3->length = new_l3;
/* IPv6 UDP checksum is mandatory */
udp0->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b0,
ip6_0, &bogus);
if (udp0->checksum == 0)
udp0->checksum = 0xffff;
udp1->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b1,
ip6_1, &bogus);
if (udp1->checksum == 0)
udp1->checksum = 0xffff;
udp2->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b2,
ip6_2, &bogus);
if (udp2->checksum == 0)
udp2->checksum = 0xffff;
udp3->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b3,
ip6_3, &bogus);
if (udp3->checksum == 0)
udp3->checksum = 0xffff;
}
pkts_encapsulated += 4;
len0 = vlib_buffer_length_in_chain (vm, b0);
len1 = vlib_buffer_length_in_chain (vm, b1);
len2 = vlib_buffer_length_in_chain (vm, b2);
len3 = vlib_buffer_length_in_chain (vm, b3);
stats_n_packets += 4;
stats_n_bytes += len0 + len1 + len2 + len3;
/* Batch stats increment on the same gtpu tunnel so counter is not
incremented per packet. Note stats are still incremented for deleted
and admin-down tunnel where packets are dropped. It is not worthwhile
to check for this rare case and affect normal path performance. */
if (PREDICT_FALSE ((sw_if_index0 != stats_sw_if_index) ||
(sw_if_index1 != stats_sw_if_index) ||
(sw_if_index2 != stats_sw_if_index) ||
(sw_if_index3 != stats_sw_if_index) ))
{
stats_n_packets -= 4;
stats_n_bytes -= len0 + len1 + len2 + len3;
if ( (sw_if_index0 == sw_if_index1 ) &&
(sw_if_index1 == sw_if_index2 ) &&
(sw_if_index2 == sw_if_index3 ) )
{
if (stats_n_packets)
vlib_increment_combined_counter
(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
thread_index, stats_sw_if_index,
stats_n_packets, stats_n_bytes);
stats_sw_if_index = sw_if_index0;
stats_n_packets = 4;
stats_n_bytes = len0 + len1 + len2 + len3;
}
else
{
vlib_increment_combined_counter
(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
thread_index, sw_if_index0, 1, len0);
vlib_increment_combined_counter
(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
thread_index, sw_if_index1, 1, len1);
vlib_increment_combined_counter
(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
thread_index, sw_if_index2, 1, len2);
vlib_increment_combined_counter
(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
thread_index, sw_if_index3, 1, len3);
}
}
if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
{
gtpu_encap_trace_t *tr =
vlib_add_trace (vm, node, b0, sizeof (*tr));
tr->tunnel_index = t0 - gtm->tunnels;
tr->teid = t0->teid;
}
if (PREDICT_FALSE(b1->flags & VLIB_BUFFER_IS_TRACED))
{
gtpu_encap_trace_t *tr =
vlib_add_trace (vm, node, b1, sizeof (*tr));
tr->tunnel_index = t1 - gtm->tunnels;
tr->teid = t1->teid;
}
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;
vlib_buffer_t * b0;
u32 flow_hash0;
u32 len0;
ip4_header_t * ip4_0;
ip6_header_t * ip6_0;
udp_header_t * udp0;
gtpu_header_t * gtpu0;
u64 * copy_src0, * copy_dst0;
u32 * copy_src_last0, * copy_dst_last0;
u16 new_l0;
ip_csum_t sum0;
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);
flow_hash0 = vnet_l2_compute_flow_hash(b0);
/* Get next node index and adj index from tunnel next_dpo */
sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_TX];
hi0 = vnet_get_sup_hw_interface (vnm, sw_if_index0);
t0 = >m->tunnels[hi0->dev_instance];
/* Note: change to always set next0 if it may be set to drop */
next0 = t0->next_dpo.dpoi_next_node;
vnet_buffer(b0)->ip.adj_index[VLIB_TX] = t0->next_dpo.dpoi_index;
/* Apply the rewrite string. $$$$ vnet_rewrite? */
vlib_buffer_advance (b0, -(word)_vec_len(t0->rewrite));
if (is_ip4)
{
ip4_0 = vlib_buffer_get_current(b0);
/* Copy the fixed header */
copy_dst0 = (u64 *) ip4_0;
copy_src0 = (u64 *) t0->rewrite;
/* Copy first 32 octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
foreach_fixed_header4_offset;
#undef _
/* Last 4 octets. Hopefully gcc will be our friend */
copy_dst_last0 = (u32 *)(©_dst0[4]);
copy_src_last0 = (u32 *)(©_src0[4]);
copy_dst_last0[0] = copy_src_last0[0];
/* Fix the IP4 checksum and length */
sum0 = ip4_0->checksum;
new_l0 = /* old_l0 always 0, see the rewrite setup */
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0));
sum0 = ip_csum_update (sum0, old_l0, new_l0, ip4_header_t,
length /* changed member */);
ip4_0->checksum = ip_csum_fold (sum0);
ip4_0->length = new_l0;
/* Fix UDP length and set source port */
udp0 = (udp_header_t *)(ip4_0+1);
new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
- sizeof (*ip4_0));
udp0->length = new_l0;
udp0->src_port = flow_hash0;
/* Fix GTPU length */
gtpu0 = (gtpu_header_t *)(udp0+1);
new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
- sizeof (*ip4_0) - sizeof(*udp0)
- GTPU_V1_HDR_LEN);
gtpu0->length = new_l0;
}
else /* ip6 path */
{
int bogus = 0;
ip6_0 = vlib_buffer_get_current(b0);
/* Copy the fixed header */
copy_dst0 = (u64 *) ip6_0;
copy_src0 = (u64 *) t0->rewrite;
/* Copy first 56 (ip6) octets 8-bytes at a time */
#define _(offs) copy_dst0[offs] = copy_src0[offs];
foreach_fixed_header6_offset;
#undef _
/* Fix IP6 payload length */
new_l0 =
clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0)
- sizeof(*ip6_0));
ip6_0->payload_length = new_l0;
/* Fix UDP length and set source port */
udp0 = (udp_header_t *)(ip6_0+1);
udp0->length = new_l0;
udp0->src_port = flow_hash0;
/* Fix GTPU length */
gtpu0 = (gtpu_header_t *)(udp0+1);
new_l0 = clib_host_to_net_u16 (vlib_buffer_length_in_chain(vm, b0)
- sizeof (*ip4_0) - sizeof(*udp0)
- GTPU_V1_HDR_LEN);
gtpu0->length = new_l0;
/* IPv6 UDP checksum is mandatory */
udp0->checksum = ip6_tcp_udp_icmp_compute_checksum(vm, b0,
ip6_0, &bogus);
if (udp0->checksum == 0)
udp0->checksum = 0xffff;
}
pkts_encapsulated ++;
len0 = vlib_buffer_length_in_chain (vm, b0);
stats_n_packets += 1;
stats_n_bytes += len0;
/* Batch stats increment on the same gtpu tunnel so counter is not
incremented per packet. Note stats are still incremented for deleted
and admin-down tunnel where packets are dropped. It is not worthwhile
to check for this rare case and affect normal path performance. */
if (PREDICT_FALSE (sw_if_index0 != stats_sw_if_index))
{
stats_n_packets -= 1;
stats_n_bytes -= len0;
if (stats_n_packets)
vlib_increment_combined_counter
(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
thread_index, stats_sw_if_index,
stats_n_packets, stats_n_bytes);
stats_n_packets = 1;
stats_n_bytes = len0;
stats_sw_if_index = sw_if_index0;
}
if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
{
gtpu_encap_trace_t *tr =
vlib_add_trace (vm, node, b0, sizeof (*tr));
tr->tunnel_index = t0 - gtm->tunnels;
tr->teid = t0->teid;
}
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);
}
/* Do we still need this now that tunnel tx stats is kept? */
vlib_node_increment_counter (vm, node->node_index,
GTPU_ENCAP_ERROR_ENCAPSULATED,
pkts_encapsulated);
/* Increment any remaining batch stats */
if (stats_n_packets)
{
vlib_increment_combined_counter
(im->combined_sw_if_counters + VNET_INTERFACE_COUNTER_TX,
thread_index, stats_sw_if_index, stats_n_packets, stats_n_bytes);
node->runtime_data[0] = stats_sw_if_index;
}
return from_frame->n_vectors;
}
// Advances to the first beyond the current
// whose value is greater than or equal to target.
// we do linear search inside block because of delta encoding
unsigned int CompressedSet::Iterator::Advance(unsigned int target){
// if the pointer points past the end
if( PREDICT_FALSE(cursor == totalDocIdNum || totalDocIdNum <= 0)){
lastAccessedDocId = NO_MORE_DOCS;
return NO_MORE_DOCS;
}
// if the pointer points to the end
if(++cursor == totalDocIdNum){
lastAccessedDocId = NO_MORE_DOCS;
return NO_MORE_DOCS;
}
// the expected behavior is to find the first element AFTER the current cursor,
// who is equal or larger than target
if(cursor != 0 && target <= lastAccessedDocId) {
target = lastAccessedDocId + 1;
}
int iterBlockIndex = cursor >> BLOCK_SIZE_BIT;
int offset = cursor & BLOCK_SIZE_MODULO;
// if there is noComp block, check noComp block
// the next element is in currently in the last block , or
// currently not in the last block, but the target is larger than
// the last element of the last compressed block
unsigned int sizeOfCurrentNoCompBlock = set->sizeOfCurrentNoCompBlock;
size_t baseListForOnlyCompBlocksSize = set->baseListForOnlyCompBlocks.size();
if(sizeOfCurrentNoCompBlock>0) {// if there exists the last decomp block
if(iterBlockIndex == compBlockNum ||
(baseListForOnlyCompBlocksSize>0 && target > set->baseListForOnlyCompBlocks[baseListForOnlyCompBlocksSize-1])) {
offset = binarySearchForFirstElementEqualOrLargerThanTarget(&(set->currentNoCompBlock[0]), 0, sizeOfCurrentNoCompBlock-1, target);
if(offset>=0){
iterBlockIndex = compBlockNum;
lastAccessedDocId = set->currentNoCompBlock[offset];
cursor = (iterBlockIndex << BLOCK_INDEX_SHIFT_BITS) + offset;
return lastAccessedDocId;
} else {
// hy: to avoid the repeated lookup next time once it reaches the end of the sequence
cursor = totalDocIdNum;
lastAccessedDocId = NO_MORE_DOCS;
return lastAccessedDocId;
}
}
}
// if we cannot not find it in the noComp block, we check the comp blocks
if(baseListForOnlyCompBlocksSize>0 && target <= set->baseListForOnlyCompBlocks[baseListForOnlyCompBlocksSize-1]) {
// for the following cases, it must exist in one of the comp blocks since target<= the last base in the comp blocks
if(offset == 0) {
// searching the right block from the current block to the last block
#ifdef PREFIX_SUM
lastAccessedDocId = advanceToTargetInTheFollowingCompBlocksNoPostProcessing(target, iterBlockIndex);
#else
lastAccessedDocId = advanceToTargetInTheFollowingCompBlocks(target, iterBlockIndex);
#endif
return lastAccessedDocId;
} else { // offset > 0, the current block has been decompressed, so, first test the first block;and then do sth like case 2
assert(offset > 0);
if(target <= set->baseListForOnlyCompBlocks[iterBlockIndex]) {
while(offset < DEFAULT_BATCH_SIZE) {
#ifdef PREFIX_SUM
lastAccessedDocId += (iterDecompBlock[offset]);
#else
lastAccessedDocId = (iterDecompBlock[offset]);
#endif
if (lastAccessedDocId >= target) {
break;
}
offset++;
}
if (offset == DEFAULT_BATCH_SIZE) {
printf("Error case 3: Impossible, we must be able to find the target %d in the block, lastAccessedDocId: %d , baseListForOnlyCompBlocks[%d]\n",
target,lastAccessedDocId,iterBlockIndex);
}
assert(offset != DEFAULT_BATCH_SIZE);
cursor = (iterBlockIndex << BLOCK_INDEX_SHIFT_BITS) + offset;
return lastAccessedDocId;
} else { // hy: there must exist other comp blocks between the current block and noComp block since target <= baseListForOnlyCompBlocks.get(baseListForOnlyCompBlocks.size()-1)
#ifdef PREFIX_SUM
lastAccessedDocId = advanceToTargetInTheFollowingCompBlocksNoPostProcessing(target, iterBlockIndex);
#else
lastAccessedDocId = advanceToTargetInTheFollowingCompBlocks(target, iterBlockIndex);
#endif
return lastAccessedDocId;
}
}
}
lastAccessedDocId = NO_MORE_DOCS;
return lastAccessedDocId;
}
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;
}
/*
* cnat_port_alloc_static_v2
* public ipv4 address/port allocator for Static Port commands
* tries to allocate same outside port as inside port
*/
cnat_errno_t
cnat_static_port_alloc_v2 (
cnat_portmap_v2_t *pm,
port_alloc_t atype,
port_pair_t pair_type,
u32 i_ipv4_address,
u16 i_port,
u32 *index,
u32 *o_ipv4_address,
u16 *o_port,
u16 static_port_range
#ifndef NO_BULK_LOGGING
, bulk_alloc_size_t bulk_size,
int *nfv9_log_req
#endif
, u16 ip_n_to_1
)
{
u32 i, hash_value, my_index, found, max_attempts;
u16 start_bit, new_port;
cnat_portmap_v2_t *my_pm = 0;
u32 pm_len = vec_len(pm);
uword bit_test_result;
#ifndef NO_BULK_LOGGING
*nfv9_log_req = BULK_ALLOC_NOT_ATTEMPTED;
#endif
if (PREDICT_FALSE(pm_len == 0)) {
return (CNAT_NO_POOL_ANY);
}
switch (atype) {
case PORT_ALLOC_ANY:
found = 0;
/*
* Try to hash the IPv4 address to get an index value to select the pm
*/
hash_value = (i_ipv4_address & 0xffff) ^
((i_ipv4_address > 16) & 0xffff);
/*
* If pm_len <= 256, compact the hash to 8 bits
*/
if (PREDICT_TRUE(pm_len <= 256)) {
hash_value = (hash_value & 0xff) ^ ((hash_value > 8) & 0xff);
}
/*
* Ensure that the hash value is in the range 0 .. (pm_len-1)
*/
my_index = hash_value % pm_len;
for (i = 0; i < PORT_PROBE_LIMIT; i++) {
my_pm = pm + my_index;
if(PREDICT_TRUE(ip_n_to_1)) {
if(PREDICT_TRUE(my_pm->private_ip_users_count < ip_n_to_1)) {
/*
* Try to find a PM with atlest 33% free and my_port free
*/
if (PREDICT_TRUE((my_pm->inuse < ((BITS_PER_INST*2)/3)) &&
clib_bitmap_get_no_check(my_pm->bm,
i_port) == 1)
#ifndef NO_BULK_LOGGING
&& check_if_stat_alloc_ok_for_bulk(my_pm, i_port,
bulk_size,
static_port_range)
#endif
) {
found = 1;
break;
}
}
} else {
/*
* Try to find a PM with atlest 33% free and my_port free
*/
if (PREDICT_TRUE((my_pm->inuse < ((BITS_PER_INST*2)/3)) &&
clib_bitmap_get_no_check(my_pm->bm,
i_port) == 1)
#ifndef NO_BULK_LOGGING
&& check_if_stat_alloc_ok_for_bulk(my_pm, i_port,
bulk_size,
static_port_range)
#endif
) {
found = 1;
break;
}
}
my_index = (my_index + 1) % pm_len;
}
/*
* If not found do it the hard way .
* "hard" way, best-fit.
*/
if (!found) {
u32 min_inuse_any, min_inuse_myport;
u32 min_index_any, min_index_myport;
min_inuse_any = min_inuse_myport = PORTS_PER_ADDR + 1;
min_index_any = min_index_myport = ~0;
for (i = 0; i < pm_len; i++) {
my_pm = pm + i;
if(PREDICT_TRUE(ip_n_to_1)) {
if(PREDICT_TRUE(my_pm->private_ip_users_count < ip_n_to_1)) {
if (PREDICT_FALSE(my_pm->inuse < min_inuse_any)) {
min_inuse_any = my_pm->inuse;
min_index_any = my_pm - pm;
}
if (PREDICT_FALSE(my_pm->inuse < min_inuse_myport)) {
if (PREDICT_TRUE(clib_bitmap_get_no_check(
my_pm->bm,i_port) == 1)
#ifndef NO_BULK_LOGGING
&& check_if_stat_alloc_ok_for_bulk(my_pm,
i_port,bulk_size,static_port_range)
#endif
) {
min_inuse_myport = my_pm->inuse;
min_index_myport = my_pm - pm;
}
}
}
} else {
if (PREDICT_FALSE(my_pm->inuse < min_inuse_any)) {
min_inuse_any = my_pm->inuse;
min_index_any = my_pm - pm;
}
if (PREDICT_FALSE(my_pm->inuse < min_inuse_myport)) {
if (PREDICT_TRUE(clib_bitmap_get_no_check(
my_pm->bm, i_port) == 1)
#ifndef NO_BULK_LOGGING
&& check_if_stat_alloc_ok_for_bulk(my_pm, i_port,
bulk_size, static_port_range)
#endif
) {
min_inuse_myport = my_pm->inuse;
min_index_myport = my_pm - pm;
}
}
}
}
/*
* Check if we have an exactly matching PM that has
* myport free. If so use it. If no such PM is
* available, use any PM
*/
if (PREDICT_TRUE(min_inuse_myport < PORTS_PER_ADDR)) {
my_pm = pm + min_index_myport;
my_index = min_index_myport;
found = 1;
} else if (PREDICT_TRUE(min_inuse_any < PORTS_PER_ADDR)) {
my_pm = pm + min_index_any;
my_index = min_index_any;
found = 1;
}
}
if (!found) {
return (CNAT_NO_PORT_ANY);
}
break;
case PORT_ALLOC_DIRECTED:
my_index = *index;
if (PREDICT_FALSE(my_index > pm_len)) {
return (CNAT_INV_PORT_DIRECT);
}
my_pm = pm + my_index;
break;
default:
return (CNAT_ERR_PARSER);
}
/* Allocate a matching port if possible */
start_bit = i_port;
found = 0;
max_attempts = BITS_PER_INST;
#ifndef NO_BULK_LOGGING
if((BULK_ALLOC_SIZE_NONE != bulk_size) &&
(i_port >= static_port_range)) {
start_bit = (start_bit/bulk_size) * bulk_size;
max_attempts = BITS_PER_INST/bulk_size;
}
#endif /* NO_BULK_LOGGING */
for (i = 0; i < max_attempts; i++) {
#ifndef NO_BULK_LOGGING
if((BULK_ALLOC_SIZE_NONE != bulk_size) &&
(i_port >= static_port_range)) {
bit_test_result = cgn_clib_bitmap_check_if_all(my_pm->bm,
start_bit, bulk_size);
}
else
#endif /* #ifndef NO_BULK_LOGGING */
bit_test_result = clib_bitmap_get_no_check(my_pm->bm, start_bit);
if (PREDICT_TRUE(bit_test_result)) {
#ifndef NO_BULK_LOGGING
if((BULK_ALLOC_SIZE_NONE != bulk_size) &&
(i_port >= static_port_range)) {
*nfv9_log_req = start_bit;
if(i==0) new_port = i_port; /* First go */
else {
new_port = bit2port(start_bit);
if (pair_type == PORT_S_ODD && (new_port & 0x1) == 0)
new_port++;
}
found = 1;
break;
}
else {
#endif /* NO_BULK_LOGGING */
new_port = bit2port(start_bit);
if (pair_type == PORT_S_ODD) {
if ((new_port & 0x1) == 1) {
found = 1;
break;
}
} else if (pair_type == PORT_S_EVEN) {
if ((new_port & 0x1) == 0) {
found = 1;
break;
}
} else {
found = 1;
break;
}
#ifndef NO_BULK_LOGGING
}
#endif
}
#ifndef NO_BULK_LOGGING
if((BULK_ALLOC_SIZE_NONE != bulk_size) &&
(i_port >= static_port_range))
start_bit = (start_bit + bulk_size) % BITS_PER_INST;
else {
#endif /* NO_BULK_LOGGING */
start_bit = (start_bit + 1) % BITS_PER_INST;
if(PREDICT_FALSE(start_bit == 0)) {
start_bit = 1; /* Port 0 is invalid, so start from 1 */
}
#ifndef NO_BULK_LOGGING
}
#endif
} /* End of for loop */
if (!found) {
/* Port allocation failure */
if (atype == PORT_ALLOC_DIRECTED) {
return (CNAT_NOT_FOUND_DIRECT);
} else {
return (CNAT_NOT_FOUND_ANY);
}
}
/* Accounting */
cgn_clib_bitmap_clear_no_check(my_pm->bm, new_port);
(my_pm->inuse)++;
*index = my_pm - pm;
*o_ipv4_address = my_pm->ipv4_address;
*o_port = new_port;
return (CNAT_SUCCESS);
}
/*
* Try to allocate a portmap structure based on atype field
*/
cnat_portmap_v2_t *
cnat_dynamic_addr_alloc_from_pm (
cnat_portmap_v2_t *pm,
port_alloc_t atype,
u32 *index,
cnat_errno_t *err,
u16 ip_n_to_1,
u32 *rseed_ip)
{
u32 i, pm_len;
int my_index;
int min_inuse, min_index;
cnat_portmap_v2_t *my_pm = 0;
*err = CNAT_NO_POOL_ANY;
pm_len = vec_len(pm);
switch(atype) {
case PORT_ALLOC_ANY:
if (PREDICT_FALSE(pm_len == 0)) {
my_pm = 0;
*err = CNAT_NO_POOL_ANY;
goto done;
}
/* "Easy" way, first address with at least 200 free ports */
for (i = 0; i < PORT_PROBE_LIMIT; i++) {
*rseed_ip = randq1(*rseed_ip);
my_index = (*rseed_ip) % pm_len;
my_pm = pm + my_index;
if (PREDICT_FALSE(ip_n_to_1)) {
if(PREDICT_TRUE(ip_n_to_1 == 1)) {
if (PREDICT_FALSE(0 == my_pm->inuse)) {
goto done;
}
} else {
if(PREDICT_TRUE(my_pm->private_ip_users_count < ip_n_to_1)) {
if (PREDICT_FALSE(my_pm->inuse < ((BITS_PER_INST*2)/3))) {
goto done;
}
}
}
} else {
if (PREDICT_FALSE(my_pm->inuse < ((BITS_PER_INST*2)/3))) {
goto done;
}
}
}
/* "hard" way, best-fit. $$$$ Throttle complaint */
min_inuse = PORTS_PER_ADDR + 1;
min_index = ~0;
for (i = 0; i < pm_len; i++) {
my_pm = pm + i;
if (PREDICT_FALSE(ip_n_to_1)) {
if(PREDICT_TRUE(ip_n_to_1 == 1)) {
if (PREDICT_FALSE(!my_pm->inuse)) {
min_inuse = my_pm->inuse;
min_index = my_pm - pm;
}
} else {
if(PREDICT_TRUE(my_pm->private_ip_users_count < ip_n_to_1)) {
if (PREDICT_TRUE(my_pm->inuse < min_inuse)) {
min_inuse = my_pm->inuse;
min_index = my_pm - pm;
}
}
}
} else {
if (PREDICT_TRUE(my_pm->inuse < min_inuse)) {
min_inuse = my_pm->inuse;
min_index = my_pm - pm;
}
}
}
if (PREDICT_TRUE(min_inuse < PORTS_PER_ADDR)) {
my_pm = pm + min_index;
my_index = min_index;
goto done;
}
/* Completely out of ports */
#ifdef DEBUG_PRINTF_ENABLED
PLATFORM_DEBUG_PRINT("%s out of ports\n", __FUNCTION__);
#endif
my_pm = 0;
*err = CNAT_NO_PORT_ANY;
break;
case PORT_ALLOC_DIRECTED:
//ASSERT(*index < pm_len);
if (PREDICT_FALSE(*index > pm_len)) {
my_pm = 0;
*err = CNAT_INV_PORT_DIRECT;
goto done;
}
my_pm = pm + *index;
my_index = *index;
break;
default:
msg_spp_err("bad allocation type in cnat_port_alloc");
my_pm = 0;
*err = CNAT_ERR_PARSER;
break;
}
done:
if (PREDICT_FALSE(my_pm == NULL)) {
return (my_pm);
}
if (PREDICT_FALSE(my_pm->inuse >= BITS_PER_INST)) {
my_pm = 0;
if (atype == PORT_ALLOC_DIRECTED) {
*err = CNAT_BAD_INUSE_DIRECT;
} else {
*err = CNAT_BAD_INUSE_ANY;
}
}
return (my_pm);
}
always_inline uword
bier_disp_dispatch_inline (vlib_main_t * vm,
vlib_node_runtime_t * node,
vlib_frame_t * from_frame)
{
u32 n_left_from, next_index, * from, * to_next;
from = vlib_frame_vector_args (from_frame);
n_left_from = from_frame->n_vectors;
next_index = node->cached_next_index;
while (n_left_from > 0)
{
u32 n_left_to_next;
vlib_get_next_frame(vm, node, next_index, to_next, n_left_to_next);
while (n_left_from > 0 && n_left_to_next > 0)
{
bier_hdr_proto_id_t pproto0;
bier_disp_entry_t *bde0;
u32 next0, bi0, bdei0;
const dpo_id_t *dpo0;
vlib_buffer_t * b0;
bier_hdr_t *hdr0;
u32 entropy0;
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);
bdei0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX];
hdr0 = vlib_buffer_get_current(b0);
bde0 = bier_disp_entry_get(bdei0);
vnet_buffer(b0)->ip.adj_index[VLIB_RX] = BIER_RX_ITF;
/*
* header is in network order - flip it, we are about to
* consume it anyway
*/
bier_hdr_ntoh(hdr0);
pproto0 = bier_hdr_get_proto_id(hdr0);
entropy0 = bier_hdr_get_entropy(hdr0);
/*
* strip the header and copy the entropy value into
* the packets flow-hash field
* DSCP mumble mumble...
*/
vlib_buffer_advance(b0, (vnet_buffer(b0)->mpls.bier.n_bytes +
sizeof(*hdr0)));
vnet_buffer(b0)->ip.flow_hash = entropy0;
/*
* use the payload proto to dispatch to the
* correct stacked DPO.
*/
dpo0 = &bde0->bde_fwd[pproto0].bde_dpo;
next0 = dpo0->dpoi_next_node;
vnet_buffer(b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index;
vnet_buffer(b0)->ip.rpf_id = bde0->bde_fwd[pproto0].bde_rpf_id;
if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
{
bier_disp_dispatch_trace_t *tr =
vlib_add_trace (vm, node, b0, sizeof (*tr));
tr->pproto = pproto0;
tr->rpf_id = vnet_buffer(b0)->ip.rpf_id;
}
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 from_frame->n_vectors;
}
always_inline uword
bier_imp_dpo_inline (vlib_main_t * vm,
vlib_node_runtime_t * node,
vlib_frame_t * from_frame,
fib_protocol_t fproto,
bier_hdr_proto_id_t bproto)
{
u32 n_left_from, next_index, * from, * to_next;
from = vlib_frame_vector_args (from_frame);
n_left_from = from_frame->n_vectors;
next_index = node->cached_next_index;
while (n_left_from > 0)
{
u32 n_left_to_next;
vlib_get_next_frame(vm, node, next_index, to_next, n_left_to_next);
while (n_left_from > 0 && n_left_to_next > 0)
{
vlib_buffer_t * b0;
bier_imp_t *bimp0;
bier_hdr_t *hdr0;
u32 bi0, bii0;
u32 next0;
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);
bii0 = vnet_buffer(b0)->ip.adj_index[VLIB_TX];
bimp0 = bier_imp_get(bii0);
if (FIB_PROTOCOL_IP4 == fproto)
{
/*
* decrement the TTL on ingress to the BIER domain
*/
ip4_header_t * ip0 = vlib_buffer_get_current(b0);
u32 checksum0;
checksum0 = ip0->checksum + clib_host_to_net_u16 (0x0100);
checksum0 += checksum0 >= 0xffff;
ip0->checksum = checksum0;
ip0->ttl -= 1;
/*
* calculate an entropy
*/
if (0 == vnet_buffer(b0)->ip.flow_hash)
{
vnet_buffer(b0)->ip.flow_hash =
ip4_compute_flow_hash (ip0, IP_FLOW_HASH_DEFAULT);
}
}
if (FIB_PROTOCOL_IP6 == fproto)
{
/*
* decrement the TTL on ingress to the BIER domain
*/
ip6_header_t * ip0 = vlib_buffer_get_current(b0);
ip0->hop_limit -= 1;
/*
* calculate an entropy
*/
if (0 == vnet_buffer(b0)->ip.flow_hash)
{
vnet_buffer(b0)->ip.flow_hash =
ip6_compute_flow_hash (ip0, IP_FLOW_HASH_DEFAULT);
}
}
/* Paint the BIER header */
vlib_buffer_advance(b0, -(sizeof(bier_hdr_t) +
bier_hdr_len_id_to_num_bytes(bimp0->bi_tbl.bti_hdr_len)));
hdr0 = vlib_buffer_get_current(b0);
/* RPF check */
if (PREDICT_FALSE(BIER_RX_ITF == vnet_buffer(b0)->ip.adj_index[VLIB_RX]))
{
next0 = 0;
}
else
{
clib_memcpy_fast(hdr0, &bimp0->bi_hdr,
(sizeof(bier_hdr_t) +
bier_hdr_len_id_to_num_bytes(bimp0->bi_tbl.bti_hdr_len)));
/*
* Fixup the entropy and protocol, both of which have a
* zero value post the paint job
*/
hdr0->bh_oam_dscp_proto |=
clib_host_to_net_u16(bproto << BIER_HDR_PROTO_FIELD_SHIFT);
hdr0->bh_first_word |=
clib_host_to_net_u32((vnet_buffer(b0)->ip.flow_hash &
BIER_HDR_ENTROPY_FIELD_MASK) <<
BIER_HDR_ENTROPY_FIELD_SHIFT);
/*
* use TTL 64 for the post enacp MPLS label/BIFT-ID
* this we be decremeted in bier_output node.
*/
vnet_buffer(b0)->mpls.ttl = 65;
/* next node */
next0 = bimp0->bi_dpo[fproto].dpoi_next_node;
vnet_buffer(b0)->ip.adj_index[VLIB_TX] =
bimp0->bi_dpo[fproto].dpoi_index;
}
if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
{
bier_imp_trace_t *tr =
vlib_add_trace (vm, node, b0, sizeof (*tr));
tr->imp = bii0;
tr->hdr = *hdr0;
}
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 from_frame->n_vectors;
}
/*
* cnat_alloc_port_from_pm
* Given a portmap structure find port/port_pair that are free
*
* The assumption in this function is that bit in bm corresponds
* to a port number. This is TRUE and hence there is no call
* to the function bit2port here, though it is done in other
* places in this file.
*
*/
static u32
cnat_alloc_port_from_pm (
u32 start_port,
u32 end_port,
cnat_portmap_v2_t *my_pm,
port_pair_t pair_type
#ifndef NO_BULK_LOGGING
, bulk_alloc_size_t bulk_size,
int *nfv9_log_req
#endif /* #ifnded NO_BULK_ALLOCATION */
)
{
u32 i;
u32 start_bit;
u32 total_ports = end_port - start_port + 1;
uword bit_test_result;
uword max_trys_to_find_port;
rseed_port = randq1(rseed_port);
start_bit = rseed_port % total_ports;
start_bit = start_bit + start_port;
#ifndef NO_BULK_LOGGING
*nfv9_log_req = BULK_ALLOC_NOT_ATTEMPTED;
if(BULK_ALLOC_SIZE_NONE != bulk_size)
{
/* We need the start port of the range to be alined on integer multiple
* of bulk_size */
max_trys_to_find_port = total_ports/bulk_size;
start_bit= ((start_bit + bulk_size -1)/bulk_size) * bulk_size;
}
else
#endif /* #ifndef NO_BULK_LOGGING */
max_trys_to_find_port = total_ports;
/* Allocate a random port / port-pair */
for (i = 0; i < max_trys_to_find_port; i++) {
/* start_bit is only a u16.. so it can rollover and become zero */
if (PREDICT_FALSE((start_bit >= end_port) ||
(start_bit < start_port))) {
start_bit = start_port;
#ifndef NO_BULK_LOGGING
if(BULK_ALLOC_SIZE_NONE != bulk_size) {
start_bit= ((start_bit + bulk_size -1)/bulk_size) * bulk_size;
}
#endif /* #ifndef NO_BULK_LOGGING */
}
/* Scan forward from random position */
#ifndef NO_BULK_LOGGING
if(BULK_ALLOC_SIZE_NONE != bulk_size) {
bit_test_result = cgn_clib_bitmap_check_if_all(my_pm->bm,
start_bit, bulk_size);
}
else
#endif /* #ifndef NO_BULK_LOGGING */
bit_test_result = clib_bitmap_get_no_check(my_pm->bm, start_bit);
if (PREDICT_TRUE(bit_test_result)) {
#ifndef NO_BULK_LOGGING
if(BULK_ALLOC_SIZE_NONE != bulk_size) {
/* Got the entire bulk range */
*nfv9_log_req = bit2port(start_bit);
return start_bit;
} else {
#endif /* #ifndef NO_BULK_LOGGING */
/*
* For PORT_PAIR, first port has to be Even
* subsequent port <= end_port
* subsequent port should be unallocated
*/
if ((start_bit & 0x1) ||
((start_bit + 1) > end_port) ||
(clib_bitmap_get_no_check(my_pm->bm,
(start_bit + 1)) == 0)) {
goto notfound;
}
return (start_bit);
#ifndef NO_BULK_LOGGING
}
#endif /* #ifndef NO_BULK_LOGGING */
} /* if( free port found ) */
notfound:
#ifndef NO_BULK_LOGGING
if(BULK_ALLOC_SIZE_NONE != bulk_size) {
start_bit += bulk_size;
} else
#endif /* #ifndef NO_BULK_LOGGING */
start_bit++;
}
return (BITS_PER_INST);
}
/*
* ip6_sixrd
*/
static uword
ip6_sixrd (vlib_main_t *vm,
vlib_node_runtime_t *node,
vlib_frame_t *frame)
{
u32 n_left_from, *from, next_index, *to_next, n_left_to_next;
vlib_node_runtime_t *error_node = vlib_node_get_runtime(vm, ip6_sixrd_node.index);
u32 encap = 0;
from = vlib_frame_vector_args(frame);
n_left_from = frame->n_vectors;
next_index = node->cached_next_index;
while (n_left_from > 0) {
vlib_get_next_frame(vm, node, next_index, to_next, n_left_to_next);
while (n_left_from > 0 && n_left_to_next > 0) {
u32 pi0;
vlib_buffer_t *p0;
sixrd_domain_t *d0;
u8 error0 = SIXRD_ERROR_NONE;
ip6_header_t *ip60;
ip4_header_t *ip4h0;
u32 next0 = IP6_SIXRD_NEXT_IP4_LOOKUP;
u32 sixrd_domain_index0 = ~0;
pi0 = to_next[0] = from[0];
from += 1;
n_left_from -= 1;
to_next +=1;
n_left_to_next -= 1;
p0 = vlib_get_buffer(vm, pi0);
ip60 = vlib_buffer_get_current(p0);
// p0->current_length = clib_net_to_host_u16(ip40->length);
d0 = ip6_sixrd_get_domain(vnet_buffer(p0)->ip.adj_index[VLIB_TX], &sixrd_domain_index0);
ASSERT(d0);
/* SIXRD calc */
u64 dal60 = clib_net_to_host_u64(ip60->dst_address.as_u64[0]);
u32 da40 = sixrd_get_addr(d0, dal60);
u16 len = clib_net_to_host_u16(ip60->payload_length) + 60;
if (da40 == 0) error0 = SIXRD_ERROR_UNKNOWN;
/* construct ipv4 header */
vlib_buffer_advance(p0, - (sizeof(ip4_header_t)));
ip4h0 = vlib_buffer_get_current(p0);
vnet_buffer(p0)->sw_if_index[VLIB_TX] = (u32)~0;
ip4h0->ip_version_and_header_length = 0x45;
ip4h0->tos = 0;
ip4h0->length = clib_host_to_net_u16(len);
ip4h0->fragment_id = 0;
ip4h0->flags_and_fragment_offset = 0;
ip4h0->ttl = 0x40;
ip4h0->protocol = IP_PROTOCOL_IPV6;
ip4h0->src_address = d0->ip4_src;
ip4h0->dst_address.as_u32 = clib_host_to_net_u32(da40);
ip4h0->checksum = ip4_header_checksum(ip4h0);
next0 = error0 == SIXRD_ERROR_NONE ? IP6_SIXRD_NEXT_IP4_LOOKUP : IP6_SIXRD_NEXT_DROP;
if (PREDICT_FALSE(p0->flags & VLIB_BUFFER_IS_TRACED)) {
sixrd_trace_t *tr = vlib_add_trace(vm, node, p0, sizeof(*tr));
tr->sixrd_domain_index = sixrd_domain_index0;
}
p0->error = error_node->errors[error0];
if (PREDICT_TRUE(error0 == SIXRD_ERROR_NONE)) encap++;
vlib_validate_buffer_enqueue_x1(vm, node, next_index, to_next, n_left_to_next, pi0, next0);
}
vlib_put_next_frame(vm, node, next_index, n_left_to_next);
}
vlib_node_increment_counter(vm, ip6_sixrd_node.index, SIXRD_ERROR_ENCAPSULATED, encap);
return frame->n_vectors;
}
cnat_errno_t
cnat_dynamic_port_alloc_rtsp (
cnat_portmap_v2_t *pm,
port_alloc_t atype,
port_pair_t pair_type,
u16 start_range,
u16 end_range,
u32 *index,
u32 *o_ipv4_address,
u16 *o_port
#ifndef NO_BULK_LOGGING
, bulk_alloc_size_t bulk_size,
int *nfv9_log_req
#endif
, u32 *rseed_ip
)
{
u32 current_timestamp;
cnat_errno_t my_err = CNAT_NO_POOL_ANY;
cnat_portmap_v2_t *my_pm = 0;
u32 alloc_bit;
ASSERT(index);
ASSERT(o_ipv4_address);
ASSERT(o_port);
my_pm = cnat_dynamic_addr_alloc_from_pm(pm, atype, index, &my_err, 0,rseed_ip);
if (PREDICT_FALSE(my_pm == NULL)) {
return (my_err);
}
#if DEBUG > 1
PLATFORM_DEBUG_PRINT("ALLOC_PORT_V2: My_Instance_Number %d: IP addr 0x%x, Inuse %d\n",
my_instance_number, my_pm->ipv4_address, my_pm->inuse);
#endif
alloc_bit =
cnat_alloc_port_from_pm(start_range, end_range, my_pm, pair_type
#ifndef NO_BULK_LOGGING
, bulk_size, nfv9_log_req
#endif /* #ifndef NO_BULK_LOGGING */
);
if (alloc_bit < BITS_PER_INST) {
if (pair_type == PORT_PAIR) {
/* Accounting */
cgn_clib_bitmap_clear_no_check (my_pm->bm, alloc_bit);
cgn_clib_bitmap_clear_no_check (my_pm->bm, alloc_bit+1);
(my_pm->inuse) += 2;
} else {
/* Accounting */
cgn_clib_bitmap_clear_no_check (my_pm->bm, alloc_bit);
(my_pm->inuse)++;
}
*index = my_pm - pm;
*o_ipv4_address = my_pm->ipv4_address;
*o_port = bit2port(alloc_bit);;
return (CNAT_SUCCESS);
}
/* Completely out of ports */
current_timestamp = spp_trace_log_get_unix_time_in_seconds();
if (PREDICT_FALSE((current_timestamp - my_pm->last_sent_timestamp) >
1000)) {
spp_printf(CNAT_NO_EXT_PORT_AVAILABLE, 0, NULL);
my_pm->last_sent_timestamp = current_timestamp;
}
/* Port allocation failure */
if (atype == PORT_ALLOC_DIRECTED) {
return (CNAT_NOT_FOUND_DIRECT);
} else {
return (CNAT_NOT_FOUND_ANY);
}
}
static void
lisp_gpe_increment_stats_counters (lisp_cp_main_t * lcm, ip_adjacency_t * adj,
vlib_buffer_t * b)
{
lisp_gpe_main_t *lgm = vnet_lisp_gpe_get_main ();
lisp_gpe_adjacency_t *ladj;
ip_address_t rloc;
index_t lai;
u32 si, di;
gid_address_t src, dst;
uword *feip;
ip46_address_to_ip_address (&adj->sub_type.nbr.next_hop, &rloc);
si = vnet_buffer (b)->sw_if_index[VLIB_TX];
lai = lisp_adj_find (&rloc, si);
ASSERT (INDEX_INVALID != lai);
ladj = pool_elt_at_index (lisp_adj_pool, lai);
u8 *lisp_data = (u8 *) vlib_buffer_get_current (b);
/* skip IP header */
if (is_v4_packet (lisp_data))
lisp_data += sizeof (ip4_header_t);
else
lisp_data += sizeof (ip6_header_t);
/* skip UDP header */
lisp_data += sizeof (udp_header_t);
// TODO: skip TCP?
/* skip LISP GPE header */
lisp_data += sizeof (lisp_gpe_header_t);
i16 saved_current_data = b->current_data;
b->current_data = lisp_data - b->data;
lisp_afi_e afi = lisp_afi_from_vnet_link_type (adj->ia_link);
get_src_and_dst_eids_from_buffer (lcm, b, &src, &dst, afi);
b->current_data = saved_current_data;
di = gid_dictionary_sd_lookup (&lcm->mapping_index_by_gid, &dst, &src);
if (PREDICT_FALSE (~0 == di))
{
clib_warning ("dst mapping not found (%U, %U)", format_gid_address,
&src, format_gid_address, &dst);
return;
}
feip = hash_get (lcm->fwd_entry_by_mapping_index, di);
if (PREDICT_FALSE (!feip))
return;
lisp_stats_key_t key;
clib_memset (&key, 0, sizeof (key));
key.fwd_entry_index = feip[0];
key.tunnel_index = ladj->tunnel_index;
uword *p = hash_get_mem (lgm->lisp_stats_index_by_key, &key);
ASSERT (p);
/* compute payload length starting after GPE */
u32 bytes = b->current_length - (lisp_data - b->data - b->current_data);
vlib_increment_combined_counter (&lgm->counters, vlib_get_thread_index (),
p[0], 1, bytes);
}
/*
* cnat_mapped_static_port_alloc_v2
* /
*/
cnat_errno_t
cnat_mapped_static_port_alloc_v2 (
cnat_portmap_v2_t *pm,
port_alloc_t atype,
u32 *index,
u32 ipv4_address,
u16 port
#ifndef NO_BULK_LOGGING
, int *nfv9_log_req,
bulk_alloc_size_t bulk_size
#endif
, u16 ip_n_to_1
)
{
int i;
u32 pm_len;
u16 bm_bit;
cnat_portmap_v2_t *my_pm = 0;
u32 my_index;
ASSERT(index);
/*
* Map the port to the bit in the pm bitmap structure.
* Note that we use ports from 1024..65535, so
* port number x corresponds to (x-1024) position in bitmap
*/
bm_bit = port2bit(port);
pm_len = vec_len(pm);
switch(atype) {
case PORT_ALLOC_ANY:
if (PREDICT_FALSE(pm_len == 0)) {
return (CNAT_NO_POOL_ANY);
}
/*
* Find the pm that is allocated for this translated IP address
*/
my_index = pm_len;
for (i = 0; i < pm_len; i++) {
my_pm = pm + i;
if (PREDICT_FALSE(my_pm->ipv4_address == ipv4_address)) {
my_index = i;
break;
}
}
if ((PREDICT_FALSE(my_index >= pm_len)) ||
((PREDICT_FALSE(ip_n_to_1)) && (PREDICT_TRUE(my_pm->private_ip_users_count >= ip_n_to_1)))) {
return (CNAT_NO_POOL_ANY);
}
break;
case PORT_ALLOC_DIRECTED:
if (PREDICT_FALSE(*index > pm_len)) {
return (CNAT_INV_PORT_DIRECT);
}
my_index = *index;
my_pm = pm + my_index;
if (PREDICT_FALSE(my_pm->ipv4_address != ipv4_address)) {
if (PREDICT_FALSE(global_debug_flag && CNAT_DEBUG_GLOBAL_ALL)) {
PLATFORM_DEBUG_PRINT("Delete all main db entry for that particular in ipv4 address\n");
}
return (CNAT_INV_PORT_DIRECT);
}
break;
default:
msg_spp_err("bad allocation type in cnat_port_alloc");
return (CNAT_ERR_PARSER);
}
if (PREDICT_FALSE(my_pm == NULL)) {
return (CNAT_NO_POOL_ANY);
}
/*
* Check if the port is already allocated to some other mapping
*/
if (PREDICT_FALSE(clib_bitmap_get_no_check (my_pm->bm, bm_bit) == 0)) {
return (CNAT_NO_POOL_ANY);
}
#if DEBUG > 1
PLATFORM_DEBUG_PRINT("ALLOC_PORT_V2: My_Instance_Number %d: IP addr 0x%x, Inuse %d\n",
my_instance_number, my_pm->ipv4_address, my_pm->inuse);
#endif
/*
* Indicate that the port is already allocated
*/
cgn_clib_bitmap_clear_no_check (my_pm->bm, bm_bit);
(my_pm->inuse)++;
*index = my_index;
return (CNAT_SUCCESS);
}