static void on_client_connect(int client_idx)
{
struct rte_mbuf *buffer = get_buffer();
if(!buffer) {
return;
}
unsigned char *data = rte_pktmbuf_mtod(buffer, unsigned char *);
*data = IPAUGENBLICK_NEW_IFACES;
data++;
rte_pktmbuf_data_len(buffer) = get_all_devices(data) + 1;
rte_ring_enqueue(ipaugenblick_clients[client_idx].client_ring,(void *)buffer);
buffer = get_buffer();
if(!buffer) {
return;
}
data = rte_pktmbuf_mtod(buffer, unsigned char *);
*data = IPAUGENBLICK_NEW_ADDRESSES;
data++;
rte_pktmbuf_data_len(buffer) = get_all_addresses(data) + 1;
rte_ring_enqueue(ipaugenblick_clients[client_idx].client_ring,(void *)buffer);
buffer = get_buffer();
if(!buffer) {
return;
}
data = rte_pktmbuf_mtod(buffer, unsigned char *);
*data = IPAUGENBLICK_END_OF_RECORD;
rte_ring_enqueue(ipaugenblick_clients[client_idx].client_ring,(void *)buffer);
}
int
urdma_accl_post_recvv(struct ibv_qp *ib_qp, const struct iovec *iov, size_t iov_size,
void *context)
{
struct usiw_recv_wqe *wqe;
struct usiw_qp *qp;
unsigned int y;
int x;
if (iov_size > DPDK_VERBS_IOV_LEN_MAX) {
return -EINVAL;
}
qp = container_of(ib_qp, struct usiw_qp, ib_qp);
x = qp_get_next_recv_wqe(qp, &wqe);
if (x < 0)
return x;
wqe->wr_context = context;
memcpy(wqe->iov, iov, iov_size * sizeof(*iov));
wqe->iov_count = iov_size;
wqe->total_request_size = 0;
for (y = 0; y < iov_size; ++y) {
wqe->total_request_size += iov[y].iov_len;
}
wqe->msn = 0;
wqe->recv_size = 0;
wqe->input_size = 0;
wqe->complete = false;
x = rte_ring_enqueue(qp->rq0.ring, wqe);
assert(x == 0);
return 0;
} /* urdma_accl_post_recvv */
void ipDeFragment(void * handle, struct ip * iphead,struct sk_buff *skb){
printf("1\n");
IpImpl * impl = (IpImpl *)handle;
int index = addrtoHash( iphead->ip_src, iphead->ip_dst);
int offset = ntohs(iphead ->ip_off);
int flags = offset&~IP_OFFSET;
offset &= IP_OFFSET;
if(((flags & IP_MF) ==0)&&(offset ==0)){// no fragment.
//printf("No fragment.\n");
struct ring_buf * ptr = (struct ring_buf *)rte_malloc("rp",sizeof(struct ring_buf *),0);
if(ptr ==NULL)OUTOFMEM
ptr -> type = 0;
ptr -> ptr = iphead;
rte_ring_enqueue(impl -> r, ptr);
}
else
{
printf("Fragment in %d.\n",index);
fflush(stdout);
addToHashTable(handle, &impl -> tables[index], iphead, skb);
}
// tables[index].addr->packets->ipFra->info.ipHead = iphead;
/*here need to add ip packet info */
/*to do :add ipFragment head*/
}
int anscli_ring_send(void *buff, int buff_len)
{
void *msg;
if(buff_len > ANS_RING_MSG_SIZE)
{
printf("Too long message size, max is %d \n", ANS_RING_MSG_SIZE);
return ANS_EMSGPOOL;
}
if (rte_mempool_get(anscli_message_pool, &msg) < 0)
{
printf("Getting message failed \n");
return ANS_EMSGPOOL;
}
rte_memcpy(msg, buff, buff_len);
if (rte_ring_enqueue(anscli_ring_tx, msg) < 0)
{
printf("Sending message to ANS stack failed \n");
rte_mempool_put(anscli_message_pool, msg);
return ANS_EMSGPOOL;
}
return 0;
}
int netdpcmd_ring_send(void *buff, int buff_len)
{
void *msg;
if(buff_len > NETDP_RING_MSG_SIZE)
{
printf("Too long message size, max is %d \n", NETDP_RING_MSG_SIZE);
return NETDP_EMSGPOOL;
}
if (rte_mempool_get(netdpcmd_message_pool, &msg) < 0)
{
printf("Getting message failed \n");
return NETDP_EMSGPOOL;
}
rte_memcpy(msg, buff, buff_len);
if (rte_ring_enqueue(netdpcmd_ring_tx, msg) < 0)
{
printf("Sending message to NETDP stack failed \n");
rte_mempool_put(netdpcmd_message_pool, msg);
return NETDP_EMSGPOOL;
}
return 0;
}
/*
* Return a buffered packet.
*/
int
onvm_nf_return_pkt(struct rte_mbuf* pkt) {
/* FIXME: should we get a batch of buffered packets and then enqueue? Can we keep stats? */
if(unlikely(rte_ring_enqueue(tx_ring, pkt) == -ENOBUFS)) {
rte_pktmbuf_free(pkt);
tx_stats->tx_drop[nf_info->instance_id]++;
return -ENOBUFS;
}
else tx_stats->tx_returned[nf_info->instance_id]++;
return 0;
}
int
urdma_accl_post_sendv(struct ibv_qp *ib_qp, struct iovec *iov, size_t iov_size,
struct urdma_ah *ah, void *context)
{
struct usiw_qp *qp;
struct ee_state *ee;
struct usiw_send_wqe *wqe;
unsigned int y;
int x;
qp = container_of(ib_qp, struct usiw_qp, ib_qp);
if (!ah && !qp_connected(qp)) {
return -EINVAL;
}
if (iov_size > DPDK_VERBS_IOV_LEN_MAX) {
return -EINVAL;
}
ee = &qp->remote_ep;
if (!ee) {
return -EINVAL;
}
x = qp_get_next_send_wqe(qp, &wqe);
if (x < 0)
return x;
wqe->opcode = usiw_wr_send;
wqe->wr_context = context;
memcpy(wqe->iov, iov, iov_size * sizeof(*iov));
wqe->iov_count = iov_size;
wqe->remote_ep = ee;
wqe->state = SEND_WQE_INIT;
wqe->msn = 0; /* will be assigned at send time */
wqe->total_length = 0;
for (y = 0; y < iov_size; ++y) {
wqe->total_length += wqe->iov[y].iov_len;
}
wqe->bytes_sent = 0;
wqe->bytes_acked = 0;
x = rte_ring_enqueue(qp->sq.ring, wqe);
assert(x == 0);
return 0;
} /* urdma_accl_post_sendv */
int socket_connect(int identifier, struct sock_addr *client_addr)
{
/* using static ip for current. furute get ip from conf*/
int i;
uint8_t ip[4];
ip[0] = 192;
ip[1] = 168;
ip[2] = 78;
ip[3] = 2;
uint32_t DestIp = 0;
static uint16_t SrcPorts = 0;
if(SrcPorts == 0) {
SrcPorts = 10000;
}
SrcPorts ++;
for(i=0; i<4; i++) {
DestIp |= ip[i] << i*8;
}
printf("opening connection connect call\n");
Socket_Send_Msg *Msg = NULL;
struct tcb *ptcb = get_tcb_by_identifier(identifier);
if (rte_mempool_get(buffer_message_pool,(void **) &Msg) < 0) {
printf ("Failed to get message buffer\n");
/// / put assert ;
}
Msg->m_Identifier = identifier;
Msg->m_Msg_Type = CONNECTION_OPEN;
if (rte_ring_enqueue(ptcb->tcb_socket_ring_send, Msg) < 0) {
printf("Failed to send message - message discarded\n");
rte_mempool_put(buffer_message_pool, Msg);
}
ptcb->ipv4_src = htonl(client_addr->ip);
ptcb->sport = client_addr->port;
ptcb->ipv4_dst = DestIp;
ptcb->dport = SrcPorts;
ptcb->next_seq = 1;
pthread_mutex_lock(&(ptcb->mutex));
ptcb->WaitingOnConnect = 1;
pthread_cond_wait(&(ptcb->condAccept), &(ptcb->mutex));
ptcb->WaitingOnConnect = 0;
pthread_mutex_unlock(&(ptcb->mutex));
// wait on sema event of syn-ack.
// remove_tcb(identifier);
return 0;
}
int
EnqueueMBuf(struct rte_mbuf *mbuf)
{
struct rte_mbuf **Msg;
if (rte_mempool_get(buffer_message_pool, (void **)&Msg) < 0) {
printf ("Failed to get rte_mbuf message buffer\n");
/// / put assert ;
return -1;
}
*Msg = mbuf;
if (rte_ring_enqueue(ip_to_ether_ring_send, Msg) < 0) {
printf("Failed to send rte_mbuf message - message discarded\n");
rte_mempool_put(buffer_message_pool, Msg);
}
else {
printf("mbuf enqueue = %p\n", mbuf);
}
return 0;
}
int
socket_close(int identifier)
{
printf("closing tcb\n");
Socket_Send_Msg *Msg = NULL;
struct tcb *ptcb = get_tcb_by_identifier(identifier);
if (rte_mempool_get(buffer_message_pool, (void **)&Msg) < 0) {
printf ("Failed to get message buffer\n");
/// / put assert ;
}
Msg->m_Identifier = identifier;
Msg->m_Msg_Type = SOCKET_CLOSE;
if (rte_ring_enqueue(ptcb->tcb_socket_ring_send, Msg) < 0) {
printf("Failed to send message - message discarded\n");
rte_mempool_put(buffer_message_pool, Msg);
}
// remove_tcb(identifier);
return 0;
}
/**
* Process a completed JOB_AES_HMAC job and keep processing jobs until
* get_completed_job return NULL
*
* @param qp Queue Pair to process
* @param job JOB_AES_HMAC job
*
* @return
* - Number of processed jobs
*/
static unsigned
handle_completed_jobs(struct aesni_mb_qp *qp, JOB_AES_HMAC *job)
{
struct rte_mbuf *m = NULL;
unsigned processed_jobs = 0;
while (job) {
processed_jobs++;
m = post_process_mb_job(qp, job);
if (m)
rte_ring_enqueue(qp->processed_pkts, (void *)m);
else
qp->qp_stats.dequeue_err_count++;
job = (*qp->ops->job.get_completed_job)(&qp->mb_mgr);
}
return processed_jobs;
}
/* TCP */
int ipaugenblick_receive(int sock,void **pbuffer,int *total_len,int *first_segment_len,void **pdesc)
{
struct rte_mbuf *mbuf;
ipaugenblick_stats_receive_called++;
/* first try to look shadow. shadow pointer saved when last mbuf delievered partially */
mbuf = ipaugenblick_get_from_shadow(sock);
if((mbuf)&&(*total_len > 0)) { /* total_len > 0 means user restricts total read count */
// printf("%s %d %d\n",__FILE__,__LINE__,*total_len);
int total_len2 = *total_len;
/* now find mbuf (if any) to be delievered partially and save it to shadown */
ipaugenblick_try_read_exact_amount(mbuf,sock,&total_len2,first_segment_len);
*pbuffer = rte_pktmbuf_mtod(mbuf,void *);
*pdesc = mbuf;
if((total_len2 > 0)&&(total_len2 < *total_len)) { /* read less than user requested, try ring */
// printf("%s %d %d\n",__FILE__,__LINE__,total_len2);
struct rte_mbuf *mbuf2 = ipaugenblick_dequeue_rx_buf(sock);
if(!mbuf2) { /* ring is empty */
*total_len = total_len2;
// printf("%s %d\n",__FILE__,__LINE__);
}
else { /* now try to find an mbuf to be delievered partially in the chain */
int total_len3 = *total_len - total_len2;
int first_segment_len_dummy;
ipaugenblick_try_read_exact_amount(mbuf2,sock,&total_len3,&first_segment_len_dummy);
struct rte_mbuf *last_mbuf = rte_pktmbuf_lastseg(mbuf);
last_mbuf->next = mbuf2;
*total_len = total_len2 + total_len3;
// printf("%s %d %d\n",__FILE__,__LINE__,total_len3);
}
}
else {
// printf("%s %d %d\n",__FILE__,__LINE__,*total_len);
//goto read_from_ring;
}
if(local_socket_descriptors[sock].shadow) {
uint32_t ringidx_ready_mask = sock|(SOCKET_READABLE_BIT << SOCKET_READY_SHIFT);
if(local_socket_descriptors[sock].select != -1)
rte_ring_enqueue(selectors[local_socket_descriptors[sock].select].ready_connections,(void *)ringidx_ready_mask);
}
return 0;
}
int
urdma_accl_post_read(struct ibv_qp *ib_qp, void *addr, size_t length,
struct urdma_ah *ah, uint64_t remote_addr, uint32_t rkey,
void *context)
{
struct usiw_send_wqe *wqe;
struct ee_state *ee;
struct usiw_qp *qp;
int x;
qp = container_of(ib_qp, struct usiw_qp, ib_qp);
if (!ah && !qp_connected(qp)) {
return -EINVAL;
}
ee = &qp->remote_ep;
if (!ee) {
return -EINVAL;
}
x = qp_get_next_send_wqe(qp, &wqe);
if (x < 0)
return x;
wqe->opcode = usiw_wr_read;
wqe->wr_context = context;
wqe->iov[0].iov_base = addr;
wqe->iov[0].iov_len = length;
wqe->iov_count = 1;
wqe->remote_addr = remote_addr;
wqe->rkey = rkey;
wqe->remote_ep = ee;
wqe->state = SEND_WQE_INIT;
wqe->msn = 0; /* will be assigned at send time */
wqe->local_stag = 0;
wqe->total_length = length;
wqe->bytes_sent = 0;
x = rte_ring_enqueue(qp->sq.ring, wqe);
assert(x == 0);
return 0;
} /* urdma_accl_post_read */
int
socket_send(int ser_id, const unsigned char *message, int len)
{
Socket_Send_Msg *Msg = NULL;
struct tcb *ptcb = get_tcb_by_identifier(ser_id);
if (rte_mempool_get(buffer_message_pool,(void **) &Msg) < 0) {
printf ("Failed to get message buffer\n");
/// / put assert ;
}
Msg->m_Identifier = ser_id;
Msg->m_Len = len;
Msg->m_Msg_Type = SEND_DATA;
memcpy(Msg->m_Data, message, len);
if (rte_ring_enqueue(ptcb->tcb_socket_ring_send, Msg) < 0) {
printf("Failed to send message - message discarded\n");
rte_mempool_put(buffer_message_pool, Msg);
}
printf("****** Enqued for %s and len %d and identifier %d\n",(char *)Msg->m_Data, Msg->m_Len, Msg->m_Identifier);
// sendtcppacket(ptcb, mbuf, message, len);
// ptcb->send_data(message, len);
return 0;
}
/**
* CALLED BY NF:
* Application main function - loops through
* receiving and processing packets. Never returns
*/
int
onvm_nf_run(struct onvm_nf_info* info, int(*handler)(struct rte_mbuf* pkt, struct onvm_pkt_meta* meta)) {
void *pkts[PKT_READ_SIZE];
struct onvm_pkt_meta* meta;
printf("\nClient process %d handling packets\n", info->instance_id);
printf("[Press Ctrl-C to quit ...]\n");
/* Listen for ^C so we can exit gracefully */
signal(SIGINT, handle_signal);
for (; keep_running;) {
uint16_t i, j, nb_pkts = PKT_READ_SIZE;
void *pktsTX[PKT_READ_SIZE];
int tx_batch_size = 0;
int ret_act;
/* try dequeuing max possible packets first, if that fails, get the
* most we can. Loop body should only execute once, maximum */
while (nb_pkts > 0 &&
unlikely(rte_ring_dequeue_bulk(rx_ring, pkts, nb_pkts) != 0))
nb_pkts = (uint16_t)RTE_MIN(rte_ring_count(rx_ring), PKT_READ_SIZE);
if(nb_pkts == 0) {
continue;
}
/* Give each packet to the user proccessing function */
for (i = 0; i < nb_pkts; i++) {
meta = onvm_get_pkt_meta((struct rte_mbuf*)pkts[i]);
ret_act = (*handler)((struct rte_mbuf*)pkts[i], meta);
/* NF returns 0 to return packets or 1 to buffer */
if(likely(ret_act == 0)) {
pktsTX[tx_batch_size++] = pkts[i];
}
else {
tx_stats->tx_buffer[info->instance_id]++;
}
}
if (unlikely(tx_batch_size > 0 && rte_ring_enqueue_bulk(tx_ring, pktsTX, tx_batch_size) == -ENOBUFS)) {
tx_stats->tx_drop[info->instance_id] += tx_batch_size;
for (j = 0; j < tx_batch_size; j++) {
rte_pktmbuf_free(pktsTX[j]);
}
} else {
tx_stats->tx[info->instance_id] += tx_batch_size;
}
}
nf_info->status = NF_STOPPED;
/* Put this NF's info struct back into queue for manager to ack shutdown */
nf_info_ring = rte_ring_lookup(_NF_QUEUE_NAME);
if (nf_info_ring == NULL) {
rte_mempool_put(nf_info_mp, nf_info); // give back mermory
rte_exit(EXIT_FAILURE, "Cannot get nf_info ring for shutdown");
}
if (rte_ring_enqueue(nf_info_ring, nf_info) < 0) {
rte_mempool_put(nf_info_mp, nf_info); // give back mermory
rte_exit(EXIT_FAILURE, "Cannot send nf_info to manager for shutdown");
}
return 0;
}
//将数据加入到分片链表中，其中，分片一定是唯一的，关于重复能不能重复这一点有待商榷
void adddToipFra(void *handle, struct srcDstAddr * fa, struct ipPacketHead * table, struct ip *iphead, struct sk_buff * skb)
{
IpImpl * impl = (IpImpl *)handle;
if (((ntohs(iphead->ip_off)&~IP_OFFSET) & IP_MF) == 0){
table->MF = 0;
}
printf("\n2 in addr\n ");
fflush(stdout);
if (table->ipFra == NULL){
table->ipFra = (struct ipFragment *)rte_malloc("ipFra", sizeof(struct ipFragment),0);
if (table ->ipFra == NULL ){printf("Out of Mem1!\n");return ;}
else{
table->ipFra->next = NULL;
table->ipFra->seq = NULL;
table->ipFra->skb = skb;
//table-> ipFra -> myJiffies = getTime();
table->fraSeq = table->ipFra;
table->ipFra->length = iphead -> ip_len;
table->ipFra->offset = ntohs(iphead->ip_off) & IP_OFFSET;
/*
if(impl -> tail == NULL){
impl -> tail = table -> ipFra;
table -> ipFra -> timer_pre = impl -> head;
impl -> head -> timer_next = table -> ipFra;
impl -> tail -> timer_next = NULL;
}else{
table -> ipFra -> timer_pre = impl -> tail;
impl -> tail -> timer_next = table -> ipFra;
impl -> tail = table -> ipFra;
impl -> tail -> timer_next = NULL;
}*/
/*
table->ipFra->timer_pre -> timer_next = table -> ipFra -> timer_next;
if(table->ipFra->timer_next)
table->ipFra->timer_next ->timer_pre = table -> ipFra -> timer_pre;
impl -> tail -> timer_next = table -> ipFra;
table -> ipFra -> timer_pre = impl -> tail;
impl -> tail = table -> ipFra;
impl -> tail -> timer_next = NULL;
*/
}
}
else{
//这里要做的是两件事
//1.记录下数据包到来的顺序。
//2.按数据包偏移位排序。
//3.记录完成后检查是否是一个完整的分片包。
//4.不完整就结束程序，完整就将包发到数据包池中。
//timer_link
/*
table -> ipFra -> myJiffies = getTime();//just change the first's myJiffies.
table->ipFra->timer_pre -> timer_next = table -> ipFra -> timer_next;
if(table->ipFra->timer_next)
table->ipFra->timer_next ->timer_pre = table -> ipFra -> timer_pre;
impl -> tail -> timer_next = table -> ipFra;
table -> ipFra -> timer_pre = impl -> tail;
impl -> tail = table -> ipFra;
impl -> tail -> timer_next = NULL;
printf("\naddr:%ld %ld %ld\n",(long)table -> ipFra, (long)table -> ipFra -> timer_pre, (long)table -> ipFra->timer_next); */
struct ipFragment * current, *pre,*newFrag;
newFrag = (struct ipFragment *)rte_malloc("Fra", sizeof(struct ipFragment),0);
if (newFrag == NULL){printf("Out of Mem2!\n");return ;}
else{
//here edit the new fragment
//to do:edit the info.
newFrag->skb = skb;
newFrag->seq = NULL;
newFrag->next = NULL;
newFrag->offset = ntohs(iphead->ip_off) & IP_OFFSET;
newFrag->length = iphead -> ip_len;
}
//1.record the sequence of packet.
current = table->fraSeq;
pre = current;
printf("3 ");
fflush(stdout);
while (current){
pre = current;
current = current->seq;
}
printf("4 ");
fflush(stdout);
pre->seq = newFrag;
//2.sort by offset
current = table->ipFra;
pre = current;
if(current -> offset == newFrag -> offset){
//if find the same packet, just change the next, not change the coming sequence.
if(current -> length < newFrag -> length)
{
newFrag -> next = current -> next;
table -> ipFra = newFrag;
}
}
else if (current->offset > newFrag->offset){
newFrag->next = current;
table->ipFra = newFrag;
}
else{
printf("5 ");
fflush(stdout);
while (current && current->offset < newFrag->offset)
{
pre = current;
current = current -> next;
if(current -> offset == newFrag -> offset){
//if find the same packet, just change the next, not change the coming sequence.
if(current -> length < newFrag -> length)
{
newFrag -> next = current -> next;
pre -> next = newFrag;
}
}
}
printf("6 ");
fflush(stdout);
pre->next = newFrag;
newFrag->next = current;
}
//3.judge weather the fragment is complete.
if (table->MF == 0){//get the last fragement, need to judge now.
//int num = 0;
printf("MF =0.\n");
current = table->ipFra;
pre = current;
printf("7 ");
fflush(stdout);
while (current -> next){
if (current->offset + current->length > current->next->offset){
pre = current;
current = current->next;
/*to do :debug!*/
printf("In loop 7.\n");
//if(pre == current){printf("The same.\n");return;}
}
else
break;
}
printf("In if.\n");
if (pre->offset + pre->length >= current->offset){
//Job done.
//here has two things to do.
//1.
//return;
struct ring_buf * ptr = (struct ring_buf *)rte_malloc("ring_buf",sizeof(struct ring_buf),0);
//void **obj = rte_malloc("rp",sizeof(void *)*2,0);
if(ptr == NULL)OUTOFMEM
ptr -> type = 1;
ptr -> ptr = table -> ipFra;
//obj[0] = ptr;
rte_ring_enqueue(impl -> r, ptr);
//printf("\naddr:%ld %ld %ld\n",(long)table -> ipFra, (long)table -> ipFra -> timer_pre, (long)table -> ipFra->timer_next);
if(impl -> tail == table )
{
//printf("3.1");
//fflush(stdout);
impl -> tail = table ->timer_pre;
impl -> tail ->timer_next = NULL;
}
else
{
//printf("3.2");
//fflush(stdout);
//printf("%ld %ld", (long)table -> ipFra -> timer_pre, (long)table -> ipFra->timer_next);
//fflush(stdout);
table->timer_pre -> timer_next = table -> timer_next;
//printf("3.3");
//fflush(stdout);
if(table -> timer_next)
table->timer_next ->timer_pre = table -> timer_pre;
}
//这边考虑如何把分片包断开
if(table -> next){
table -> next -> pre = table -> pre;
}//here just for test ring
//printf("2");
//fflush(stdout);
if(table -> pre){
table -> pre -> next = table -> next;
}else
{
fa -> packets = NULL;
}
realsePacket(handle, ptr->ptr);
//printf("3");
//fflush(stdout);
//printf("4");
//fflush(stdout);
// ptr = NULL;
// rte_ring_dequeue(impl -> r, (void **)&ptr);
//ptr = getPacket(handle);
//printf("In ring %d IP:%p.\n",ptr -> type, ptr -> ptr);
}else
printf("Job not done!\n");
//else the fragement not completed, just continue.
}
printf("8 ");
fflush(stdout);
}
}
void send_loop(void)
{
RTE_LOG(INFO, APP, "send_loop()\n");
char pkt[PKT_SIZE] = {0};
int nreceived;
int retval = 0;
(void) retval;
#ifdef CALC_CHECKSUM
unsigned int kk = 0;
#endif
srand(time(NULL));
//Initializate packet contents
int i;
for(i = 0; i < PKT_SIZE; i++)
pkt[i] = rand()%256;
#if ALLOC_METHOD == ALLOC_APP
struct rte_mempool * packets_pool = rte_mempool_lookup("ovs_mp_1500_0_262144");
//struct rte_mempool * packets_pool = rte_mempool_lookup("packets");
//Create mempool
//struct rte_mempool * packets_pool = rte_mempool_create(
// "packets",
// NUM_PKTS,
// MBUF_SIZE,
// CACHE_SIZE, //This is the size of the mempool cache
// sizeof(struct rte_pktmbuf_pool_private),
// rte_pktmbuf_pool_init,
// NULL,
// rte_pktmbuf_init,
// NULL,
// rte_socket_id(),
// 0 /*NO_FLAGS*/);
if(packets_pool == NULL)
{
RTE_LOG(INFO, APP, "rte_errno: %s\n", rte_strerror(rte_errno));
rte_exit(EXIT_FAILURE, "Cannot find memory pool\n");
}
RTE_LOG(INFO, APP, "There are %d free packets in the pool\n",
rte_mempool_count(packets_pool));
#endif
#ifdef USE_BURST
struct rte_mbuf * packets_array[BURST_SIZE] = {0};
struct rte_mbuf * packets_array_rx[BURST_SIZE] = {0};
int ntosend;
int n;
(void) n;
/* prealloc packets */
do
{
n = rte_mempool_get_bulk(packets_pool, (void **) packets_array, BURST_SIZE);
} while(n != 0 && !stop);
ntosend = BURST_SIZE;
#else
struct rte_mbuf * mbuf;
/* prealloc packet */
do {
mbuf = rte_pktmbuf_alloc(packets_pool);
} while(mbuf == NULL);
#endif
RTE_LOG(INFO, APP, "Starting sender loop\n");
signal (SIGINT, crtl_c_handler);
stop = 0;
while(likely(!stop))
{
while(pause_);
#ifdef USE_BURST
#if ALLOC_METHOD == ALLOC_OVS
//Try to get BURS_SIZE free slots
ntosend = rte_ring_dequeue_burst(alloc_q, (void **) packets_array, BURST_SIZE);
#elif ALLOC_METHOD == ALLOC_APP
//do
//{
// n = rte_mempool_get_bulk(packets_pool, (void **) packets_array, BURST_SIZE);
//} while(n != 0 && !stop);
//ntosend = BURST_SIZE;
#else
#error "No implemented"
#endif
//Copy data to the buffers
for(i = 0; i < ntosend; i++)
{
rte_memcpy(packets_array[i]->buf_addr, pkt, PKT_SIZE);
//fill_packet(packets_array[i]->pkt.data);
packets_array[i]->next = NULL;
packets_array[i]->pkt_len = PKT_SIZE;
packets_array[i]->data_len = PKT_SIZE;
#ifdef CALC_CHECKSUM
for(i = 0; i < ntosend; i++)
for(kk = 0; kk < 8; kk++)
checksum += ((uint64_t *)packets_array[i]->buf_addr)[kk];
#endif
}
//Enqueue data (try until all the allocated packets are enqueue)
i = 0;
while(i < ntosend && !stop)
{
i += rte_ring_enqueue_burst(tx_ring, (void **) &packets_array[i], ntosend - i);
/* also dequeue some packets */
nreceived= rte_ring_dequeue_burst(rx_ring, (void **) packets_array_rx, BURST_SIZE);
rx += nreceived; /* update statistics */
}
#else // [NO] USE_BURST
#if ALLOC_METHOD == ALLOC_OVS //Method 1
//Read a buffer to be used as a buffer for a packet
retval = rte_ring_dequeue(alloc_q, (void **)&mbuf);
if(retval != 0)
{
#ifdef CALC_ALLOC_STATS
//stats.alloc_fails++;
#endif
continue;
}
#elif ALLOC_METHOD == ALLOC_APP //Method 2
//mbuf = rte_pktmbuf_alloc(packets_pool);
//if(mbuf == NULL)
//{
//#ifdef CALC_ALLOC_STATS
// stats.alloc_fails++;
//#endif
// continue;
//}
#else
#error "ALLOC_METHOD has a non valid value"
#endif
#if DELAY_CYCLES > 0
//This loop increases mumber of packets per second (don't ask me why)
unsigned long long j = 0;
for(j = 0; j < DELAY_CYCLES; j++)
asm("");
#endif
//Copy packet to the correct buffer
rte_memcpy(mbuf->buf_addr, pkt, PKT_SIZE);
//fill_packet(mbuf->pkt.data);
//mbuf->pkt.next = NULL;
//mbuf->pkt.pkt_len = PKT_SIZE;
//mbuf->pkt.data_len = PKT_SIZE;
(void) pkt;
mbuf->next = NULL;
mbuf->pkt_len = PKT_SIZE;
mbuf->data_len = PKT_SIZE;
#ifdef CALC_CHECKSUM
for(kk = 0; kk < 8; kk++)
checksum += ((uint64_t *)mbuf->buf_addr)[kk];
#endif
//this method avoids dropping packets:
//Simple tries until the packet is inserted in the queue
tryagain:
retval = rte_ring_enqueue(tx_ring, (void *) mbuf);
if(retval == -ENOBUFS && !stop)
{
#ifdef CALC_TX_TRIES
//stats.tx_retries++;
#endif
goto tryagain;
}
#ifdef CALC_TX_STATS
//stats.tx++;
#endif
#endif //USE_BURST
}
#ifdef CALC_CHECKSUM
printf("Checksum was %" PRIu64 "\n", checksum);
#endif
}
/*
* Initialize a given port using default settings and with the RX buffers
* coming from the mbuf_pool passed as a parameter.
* FIXME: Starting with assumption of one thread/core per port
*/
static inline int uhd_dpdk_port_init(struct uhd_dpdk_port *port,
struct rte_mempool *rx_mbuf_pool,
unsigned int mtu)
{
int retval;
/* Check for a valid port */
if (port->id >= rte_eth_dev_count())
return -ENODEV;
/* Set up Ethernet device with defaults (1 RX ring, 1 TX ring) */
/* FIXME: Check if hw_ip_checksum is possible */
struct rte_eth_conf port_conf = {
.rxmode = {
.max_rx_pkt_len = mtu,
.jumbo_frame = 1,
.hw_ip_checksum = 1,
}
};
retval = rte_eth_dev_configure(port->id, 1, 1, &port_conf);
if (retval != 0)
return retval;
retval = rte_eth_rx_queue_setup(port->id, 0, DEFAULT_RING_SIZE,
rte_eth_dev_socket_id(port->id), NULL, rx_mbuf_pool);
if (retval < 0)
return retval;
retval = rte_eth_tx_queue_setup(port->id, 0, DEFAULT_RING_SIZE,
rte_eth_dev_socket_id(port->id), NULL);
if (retval < 0)
goto port_init_fail;
/* Create the hash table for the RX sockets */
char name[32];
snprintf(name, sizeof(name), "rx_table_%u", port->id);
struct rte_hash_parameters hash_params = {
.name = name,
.entries = UHD_DPDK_MAX_SOCKET_CNT,
.key_len = sizeof(struct uhd_dpdk_ipv4_5tuple),
.hash_func = NULL,
.hash_func_init_val = 0,
};
port->rx_table = rte_hash_create(&hash_params);
if (port->rx_table == NULL) {
retval = rte_errno;
goto port_init_fail;
}
/* Create ARP table */
snprintf(name, sizeof(name), "arp_table_%u", port->id);
hash_params.name = name;
hash_params.entries = UHD_DPDK_MAX_SOCKET_CNT;
hash_params.key_len = sizeof(uint32_t);
hash_params.hash_func = NULL;
hash_params.hash_func_init_val = 0;
port->arp_table = rte_hash_create(&hash_params);
if (port->arp_table == NULL) {
retval = rte_errno;
goto free_rx_table;
}
/* Set up list for TX queues */
LIST_INIT(&port->txq_list);
/* Start the Ethernet port. */
retval = rte_eth_dev_start(port->id);
if (retval < 0) {
goto free_arp_table;
}
/* Display the port MAC address. */
rte_eth_macaddr_get(port->id, &port->mac_addr);
RTE_LOG(INFO, EAL, "Port %u MAC: %02x %02x %02x %02x %02x %02x\n",
(unsigned)port->id,
port->mac_addr.addr_bytes[0], port->mac_addr.addr_bytes[1],
port->mac_addr.addr_bytes[2], port->mac_addr.addr_bytes[3],
port->mac_addr.addr_bytes[4], port->mac_addr.addr_bytes[5]);
struct rte_eth_link link;
rte_eth_link_get(port->id, &link);
RTE_LOG(INFO, EAL, "Port %u UP: %d\n", port->id, link.link_status);
return 0;
free_arp_table:
rte_hash_free(port->arp_table);
free_rx_table:
rte_hash_free(port->rx_table);
port_init_fail:
return rte_errno;
}
static int uhd_dpdk_thread_init(struct uhd_dpdk_thread *thread, unsigned int id)
{
if (!ctx || !thread)
return -EINVAL;
unsigned int socket_id = rte_lcore_to_socket_id(id);
thread->id = id;
thread->rx_pktbuf_pool = ctx->rx_pktbuf_pools[socket_id];
thread->tx_pktbuf_pool = ctx->tx_pktbuf_pools[socket_id];
LIST_INIT(&thread->port_list);
char name[32];
snprintf(name, sizeof(name), "sockreq_ring_%u", id);
thread->sock_req_ring = rte_ring_create(
name,
UHD_DPDK_MAX_PENDING_SOCK_REQS,
socket_id,
RING_F_SC_DEQ
);
if (!thread->sock_req_ring)
return -ENOMEM;
return 0;
}
int uhd_dpdk_init(int argc, char **argv, unsigned int num_ports,
int *port_thread_mapping, int num_mbufs, int mbuf_cache_size,
int mtu)
{
/* Init context only once */
if (ctx)
return 1;
if ((num_ports == 0) || (port_thread_mapping == NULL)) {
return -EINVAL;
}
/* Grabs arguments intended for DPDK's EAL */
int ret = rte_eal_init(argc, argv);
if (ret < 0)
rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
ctx = (struct uhd_dpdk_ctx *) rte_zmalloc("uhd_dpdk_ctx", sizeof(*ctx), rte_socket_id());
if (!ctx)
return -ENOMEM;
ctx->num_threads = rte_lcore_count();
if (ctx->num_threads <= 1)
rte_exit(EXIT_FAILURE, "Error: No worker threads enabled\n");
/* Check that we have ports to send/receive on */
ctx->num_ports = rte_eth_dev_count();
if (ctx->num_ports < 1)
rte_exit(EXIT_FAILURE, "Error: Found no ports\n");
if (ctx->num_ports < num_ports)
rte_exit(EXIT_FAILURE, "Error: User requested more ports than available\n");
/* Get memory for thread and port data structures */
ctx->threads = rte_zmalloc("uhd_dpdk_thread", RTE_MAX_LCORE*sizeof(struct uhd_dpdk_thread), 0);
if (!ctx->threads)
rte_exit(EXIT_FAILURE, "Error: Could not allocate memory for thread data\n");
ctx->ports = rte_zmalloc("uhd_dpdk_port", ctx->num_ports*sizeof(struct uhd_dpdk_port), 0);
if (!ctx->ports)
rte_exit(EXIT_FAILURE, "Error: Could not allocate memory for port data\n");
/* Initialize the thread data structures */
for (int i = rte_get_next_lcore(-1, 1, 0);
(i < RTE_MAX_LCORE);
i = rte_get_next_lcore(i, 1, 0))
{
/* Do one mempool of RX/TX per socket */
unsigned int socket_id = rte_lcore_to_socket_id(i);
/* FIXME Probably want to take into account actual number of ports per socket */
if (ctx->tx_pktbuf_pools[socket_id] == NULL) {
/* Creates a new mempool in memory to hold the mbufs.
* This is done for each CPU socket
*/
const int mbuf_size = mtu + 2048 + RTE_PKTMBUF_HEADROOM;
char name[32];
snprintf(name, sizeof(name), "rx_mbuf_pool_%u", socket_id);
ctx->rx_pktbuf_pools[socket_id] = rte_pktmbuf_pool_create(
name,
ctx->num_ports*num_mbufs,
mbuf_cache_size,
0,
mbuf_size,
socket_id
);
snprintf(name, sizeof(name), "tx_mbuf_pool_%u", socket_id);
ctx->tx_pktbuf_pools[socket_id] = rte_pktmbuf_pool_create(
name,
ctx->num_ports*num_mbufs,
mbuf_cache_size,
0,
mbuf_size,
socket_id
);
if ((ctx->rx_pktbuf_pools[socket_id]== NULL) ||
(ctx->tx_pktbuf_pools[socket_id]== NULL))
rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
}
if (uhd_dpdk_thread_init(&ctx->threads[i], i) < 0)
rte_exit(EXIT_FAILURE, "Error initializing thread %i\n", i);
}
unsigned master_lcore = rte_get_master_lcore();
/* Assign ports to threads and initialize the port data structures */
for (unsigned int i = 0; i < num_ports; i++) {
int thread_id = port_thread_mapping[i];
if (thread_id < 0)
continue;
if (((unsigned int) thread_id) == master_lcore)
RTE_LOG(WARNING, EAL, "User requested master lcore for port %u\n", i);
if (ctx->threads[thread_id].id != (unsigned int) thread_id)
rte_exit(EXIT_FAILURE, "Requested inactive lcore %u for port %u\n", (unsigned int) thread_id, i);
struct uhd_dpdk_port *port = &ctx->ports[i];
port->id = i;
port->parent = &ctx->threads[thread_id];
ctx->threads[thread_id].num_ports++;
LIST_INSERT_HEAD(&ctx->threads[thread_id].port_list, port, port_entry);
/* Initialize port. */
if (uhd_dpdk_port_init(port, port->parent->rx_pktbuf_pool, mtu) != 0)
rte_exit(EXIT_FAILURE, "Cannot init port %"PRIu8 "\n",
i);
}
RTE_LOG(INFO, EAL, "Init DONE!\n");
/* FIXME: Create functions to do this */
RTE_LOG(INFO, EAL, "Starting I/O threads!\n");
for (int i = rte_get_next_lcore(-1, 1, 0);
(i < RTE_MAX_LCORE);
i = rte_get_next_lcore(i, 1, 0))
{
struct uhd_dpdk_thread *t = &ctx->threads[i];
if (!LIST_EMPTY(&t->port_list)) {
rte_eal_remote_launch(_uhd_dpdk_driver_main, NULL, ctx->threads[i].id);
}
}
return 0;
}
/* FIXME: This will be changed once we have functions to handle the threads */
int uhd_dpdk_destroy(void)
{
if (!ctx)
return -ENODEV;
struct uhd_dpdk_config_req *req = (struct uhd_dpdk_config_req *) rte_zmalloc(NULL, sizeof(*req), 0);
if (!req)
return -ENOMEM;
req->req_type = UHD_DPDK_LCORE_TERM;
for (int i = rte_get_next_lcore(-1, 1, 0);
(i < RTE_MAX_LCORE);
i = rte_get_next_lcore(i, 1, 0))
{
struct uhd_dpdk_thread *t = &ctx->threads[i];
if (LIST_EMPTY(&t->port_list))
continue;
if (rte_eal_get_lcore_state(t->id) == FINISHED)
continue;
pthread_mutex_init(&req->mutex, NULL);
pthread_cond_init(&req->cond, NULL);
pthread_mutex_lock(&req->mutex);
if (rte_ring_enqueue(t->sock_req_ring, req)) {
pthread_mutex_unlock(&req->mutex);
RTE_LOG(ERR, USER2, "Failed to terminate thread %d\n", i);
rte_free(req);
return -ENOSPC;
}
struct timespec timeout = {
.tv_sec = 1,
.tv_nsec = 0
};
pthread_cond_timedwait(&req->cond, &req->mutex, &timeout);
pthread_mutex_unlock(&req->mutex);
}
rte_free(req);
return 0;
}
/**
* CALLED BY NF:
* Initialises everything we need
*
* Returns the number of arguments parsed by both rte_eal_init and
* parse_nflib_args offset by 1. This is used by getopt in the NF's
* code. The offsetting by one accounts for getopt parsing "--" which
* increments optind by 1 each time.
*/
int
onvm_nf_init(int argc, char *argv[], const char *nf_tag) {
const struct rte_memzone *mz;
const struct rte_memzone *mz_scp;
struct rte_mempool *mp;
struct onvm_service_chain **scp;
int retval_eal, retval_parse, retval_final;
if ((retval_eal = rte_eal_init(argc, argv)) < 0)
return -1;
/* Modify argc and argv to conform to getopt rules for parse_nflib_args */
argc -= retval_eal; argv += retval_eal;
/* Reset getopt global variables opterr and optind to their default values */
opterr = 0; optind = 1;
if ((retval_parse = parse_nflib_args(argc, argv)) < 0)
rte_exit(EXIT_FAILURE, "Invalid command-line arguments\n");
/*
* Calculate the offset that the nf will use to modify argc and argv for its
* getopt call. This is the sum of the number of arguments parsed by
* rte_eal_init and parse_nflib_args. This will be decremented by 1 to assure
* getopt is looking at the correct index since optind is incremented by 1 each
* time "--" is parsed.
* This is the value that will be returned if initialization succeeds.
*/
retval_final = (retval_eal + retval_parse) - 1;
/* Reset getopt global variables opterr and optind to their default values */
opterr = 0; optind = 1;
/* Lookup mempool for nf_info struct */
nf_info_mp = rte_mempool_lookup(_NF_MEMPOOL_NAME);
if (nf_info_mp == NULL)
rte_exit(EXIT_FAILURE, "No Client Info mempool - bye\n");
/* Initialize the info struct */
nf_info = ovnm_nf_info_init(nf_tag);
mp = rte_mempool_lookup(PKTMBUF_POOL_NAME);
if (mp == NULL)
rte_exit(EXIT_FAILURE, "Cannot get mempool for mbufs\n");
mz = rte_memzone_lookup(MZ_CLIENT_INFO);
if (mz == NULL)
rte_exit(EXIT_FAILURE, "Cannot get tx info structure\n");
tx_stats = mz->addr;
mz_scp = rte_memzone_lookup(MZ_SCP_INFO);
if (mz_scp == NULL)
rte_exit(EXIT_FAILURE, "Cannot get service chain info structre\n");
scp = mz_scp->addr;
default_chain = *scp;
onvm_sc_print(default_chain);
nf_info_ring = rte_ring_lookup(_NF_QUEUE_NAME);
if (nf_info_ring == NULL)
rte_exit(EXIT_FAILURE, "Cannot get nf_info ring");
/* Put this NF's info struct onto queue for manager to process startup */
if (rte_ring_enqueue(nf_info_ring, nf_info) < 0) {
rte_mempool_put(nf_info_mp, nf_info); // give back mermory
rte_exit(EXIT_FAILURE, "Cannot send nf_info to manager");
}
/* Wait for a client id to be assigned by the manager */
RTE_LOG(INFO, APP, "Waiting for manager to assign an ID...\n");
for (; nf_info->status == (uint16_t)NF_WAITING_FOR_ID ;) {
sleep(1);
}
/* This NF is trying to declare an ID already in use. */
if (nf_info->status == NF_ID_CONFLICT) {
rte_mempool_put(nf_info_mp, nf_info);
rte_exit(NF_ID_CONFLICT, "Selected ID already in use. Exiting...\n");
} else if(nf_info->status == NF_NO_IDS) {
rte_mempool_put(nf_info_mp, nf_info);
rte_exit(NF_NO_IDS, "There are no ids available for this NF\n");
} else if(nf_info->status != NF_STARTING) {
rte_mempool_put(nf_info_mp, nf_info);
rte_exit(EXIT_FAILURE, "Error occurred during manager initialization\n");
}
RTE_LOG(INFO, APP, "Using Instance ID %d\n", nf_info->instance_id);
RTE_LOG(INFO, APP, "Using Service ID %d\n", nf_info->service_id);
/* Now, map rx and tx rings into client space */
rx_ring = rte_ring_lookup(get_rx_queue_name(nf_info->instance_id));
if (rx_ring == NULL)
rte_exit(EXIT_FAILURE, "Cannot get RX ring - is server process running?\n");
tx_ring = rte_ring_lookup(get_tx_queue_name(nf_info->instance_id));
if (tx_ring == NULL)
rte_exit(EXIT_FAILURE, "Cannot get TX ring - is server process running?\n");
/* Tell the manager we're ready to recieve packets */
nf_info->status = NF_RUNNING;
RTE_LOG(INFO, APP, "Finished Process Init.\n");
return retval_final;
}