xref: /mOS-networking-stack/core/src/core.c (revision 3b6b9ba6)
1 #define _GNU_SOURCE
2 #include <sched.h>
3 #include <unistd.h>
4 #include <sys/time.h>
5 #include <semaphore.h>
6 #include <sys/mman.h>
7 #include <signal.h>
8 #include <assert.h>
9 #include <string.h>
10 
11 #include "cpu.h"
12 #include "eth_in.h"
13 #include "fhash.h"
14 #include "tcp_send_buffer.h"
15 #include "tcp_ring_buffer.h"
16 #include "socket.h"
17 #include "eth_out.h"
18 #include "tcp.h"
19 #include "tcp_in.h"
20 #include "tcp_out.h"
21 #include "mtcp_api.h"
22 #include "eventpoll.h"
23 #include "logger.h"
24 #include "config.h"
25 #include "arp.h"
26 #include "ip_out.h"
27 #include "timer.h"
28 #include "debug.h"
29 #include "event_callback.h"
30 #include "tcp_rb.h"
31 #include "tcp_stream.h"
32 #include "io_module.h"
33 
34 #ifdef ENABLE_DPDK
35 /* for launching rte thread */
36 #include <rte_launch.h>
37 #include <rte_lcore.h>
38 #endif /* !ENABLE_DPDK */
39 #define PS_CHUNK_SIZE 64
40 #define RX_THRESH (PS_CHUNK_SIZE * 0.8)
41 
42 #define ROUND_STAT FALSE
43 #define TIME_STAT FALSE
44 #define EVENT_STAT FALSE
45 #define TESTING FALSE
46 
47 #define LOG_FILE_NAME "log"
48 #define MAX_FILE_NAME 1024
49 
50 #define MAX(a, b) ((a)>(b)?(a):(b))
51 #define MIN(a, b) ((a)<(b)?(a):(b))
52 
53 #define PER_STREAM_SLICE 0.1		// in ms
54 #define PER_STREAM_TCHECK 1			// in ms
55 #define PS_SELECT_TIMEOUT 100		// in us
56 
57 #define GBPS(bytes) (bytes * 8.0 / (1000 * 1000 * 1000))
58 
59 /*----------------------------------------------------------------------------*/
60 /* handlers for threads */
61 struct mtcp_thread_context *g_pctx[MAX_CPUS] = {0};
62 struct log_thread_context *g_logctx[MAX_CPUS] = {0};
63 /*----------------------------------------------------------------------------*/
64 static pthread_t g_thread[MAX_CPUS] = {0};
65 static pthread_t log_thread[MAX_CPUS] = {0};
66 /*----------------------------------------------------------------------------*/
67 static sem_t g_init_sem[MAX_CPUS];
68 static sem_t g_done_sem[MAX_CPUS];
69 static int running[MAX_CPUS] = {0};
70 /*----------------------------------------------------------------------------*/
71 mtcp_sighandler_t app_signal_handler;
72 static int sigint_cnt[MAX_CPUS] = {0};
73 static struct timespec sigint_ts[MAX_CPUS];
74 /*----------------------------------------------------------------------------*/
75 #ifdef NETSTAT
76 #if NETSTAT_TOTAL
77 static int printer = -1;
78 #if ROUND_STAT
79 #endif /* ROUND_STAT */
80 #endif /* NETSTAT_TOTAL */
81 #endif /* NETSTAT */
82 /*----------------------------------------------------------------------------*/
83 void
84 HandleSignal(int signal)
85 {
86 	int i = 0;
87 
88 	if (signal == SIGINT) {
89 #ifdef DARWIN
90 		int core = 0;
91 #else
92 		int core = sched_getcpu();
93 #endif
94 		struct timespec cur_ts;
95 
96 		clock_gettime(CLOCK_REALTIME, &cur_ts);
97 
98 		if (sigint_cnt[core] > 0 && cur_ts.tv_sec == sigint_ts[core].tv_sec) {
99 			for (i = 0; i < g_config.mos->num_cores; i++) {
100 				if (running[i]) {
101 					exit(0);
102 					g_pctx[i]->exit = TRUE;
103 				}
104 			}
105 		} else {
106 			for (i = 0; i < g_config.mos->num_cores; i++) {
107 				if (g_pctx[i])
108 					g_pctx[i]->interrupt = TRUE;
109 			}
110 			if (!app_signal_handler) {
111 				for (i = 0; i < g_config.mos->num_cores; i++) {
112 					if (running[i]) {
113 						exit(0);
114 						g_pctx[i]->exit = TRUE;
115 					}
116 				}
117 			}
118 		}
119 		sigint_cnt[core]++;
120 		clock_gettime(CLOCK_REALTIME, &sigint_ts[core]);
121 	}
122 
123 	if (signal != SIGUSR1) {
124 		if (app_signal_handler) {
125 			app_signal_handler(signal);
126 		}
127 	}
128 }
129 /*----------------------------------------------------------------------------*/
130 static int
131 AttachDevice(struct mtcp_thread_context* ctx)
132 {
133 	int working = -1;
134 	mtcp_manager_t mtcp = ctx->mtcp_manager;
135 
136 	if (mtcp->iom->link_devices)
137 		working = mtcp->iom->link_devices(ctx);
138 	else
139 		return 0;
140 
141 	return working;
142 }
143 /*----------------------------------------------------------------------------*/
144 #ifdef TIMESTAT
145 static inline void
146 InitStatCounter(struct stat_counter *counter)
147 {
148 	counter->cnt = 0;
149 	counter->sum = 0;
150 	counter->max = 0;
151 	counter->min = 0;
152 }
153 /*----------------------------------------------------------------------------*/
154 static inline void
155 UpdateStatCounter(struct stat_counter *counter, int64_t value)
156 {
157 	counter->cnt++;
158 	counter->sum += value;
159 	if (value > counter->max)
160 		counter->max = value;
161 	if (counter->min == 0 || value < counter->min)
162 		counter->min = value;
163 }
164 /*----------------------------------------------------------------------------*/
165 static inline uint64_t
166 GetAverageStat(struct stat_counter *counter)
167 {
168 	return counter->cnt ? (counter->sum / counter->cnt) : 0;
169 }
170 /*----------------------------------------------------------------------------*/
171 static inline int64_t
172 TimeDiffUs(struct timeval *t2, struct timeval *t1)
173 {
174 	return (t2->tv_sec - t1->tv_sec) * 1000000 +
175 			(int64_t)(t2->tv_usec - t1->tv_usec);
176 }
177 /*----------------------------------------------------------------------------*/
178 #endif
179 #ifdef NETSTAT
180 static inline void
181 PrintThreadNetworkStats(mtcp_manager_t mtcp, struct net_stat *ns)
182 {
183 	int i;
184 
185 	for (i = 0; i < g_config.mos->netdev_table->num; i++) {
186 		ns->rx_packets[i] = mtcp->nstat.rx_packets[i] - mtcp->p_nstat.rx_packets[i];
187 		ns->rx_errors[i] = mtcp->nstat.rx_errors[i] - mtcp->p_nstat.rx_errors[i];
188 		ns->rx_bytes[i] = mtcp->nstat.rx_bytes[i] - mtcp->p_nstat.rx_bytes[i];
189 		ns->tx_packets[i] = mtcp->nstat.tx_packets[i] - mtcp->p_nstat.tx_packets[i];
190 		ns->tx_drops[i] = mtcp->nstat.tx_drops[i] - mtcp->p_nstat.tx_drops[i];
191 		ns->tx_bytes[i] = mtcp->nstat.tx_bytes[i] - mtcp->p_nstat.tx_bytes[i];
192 #if NETSTAT_PERTHREAD
193 		if (g_config.mos->netdev_table->ent[i]->stat_print) {
194 			fprintf(stderr, "[CPU%2d] %s flows: %6u, "
195 					"RX: %7llu(pps) (err: %5llu), %5.2lf(Gbps), "
196 					"TX: %7llu(pps), %5.2lf(Gbps)\n",
197 					mtcp->ctx->cpu,
198 					g_config.mos->netdev_table->ent[i]->dev_name,
199 					(unsigned)mtcp->flow_cnt,
200 					(long long unsigned)ns->rx_packets[i],
201 					(long long unsigned)ns->rx_errors[i],
202 					GBPS(ns->rx_bytes[i]),
203 					(long long unsigned)ns->tx_packets[i],
204 					GBPS(ns->tx_bytes[i]));
205 		}
206 #endif
207 	}
208 	mtcp->p_nstat = mtcp->nstat;
209 
210 }
211 /*----------------------------------------------------------------------------*/
212 #if ROUND_STAT
213 static inline void
214 PrintThreadRoundStats(mtcp_manager_t mtcp, struct run_stat *rs)
215 {
216 #define ROUND_DIV (1000)
217 	rs->rounds = mtcp->runstat.rounds - mtcp->p_runstat.rounds;
218 	rs->rounds_rx = mtcp->runstat.rounds_rx - mtcp->p_runstat.rounds_rx;
219 	rs->rounds_rx_try = mtcp->runstat.rounds_rx_try - mtcp->p_runstat.rounds_rx_try;
220 	rs->rounds_tx = mtcp->runstat.rounds_tx - mtcp->p_runstat.rounds_tx;
221 	rs->rounds_tx_try = mtcp->runstat.rounds_tx_try - mtcp->p_runstat.rounds_tx_try;
222 	rs->rounds_select = mtcp->runstat.rounds_select - mtcp->p_runstat.rounds_select;
223 	rs->rounds_select_rx = mtcp->runstat.rounds_select_rx - mtcp->p_runstat.rounds_select_rx;
224 	rs->rounds_select_tx = mtcp->runstat.rounds_select_tx - mtcp->p_runstat.rounds_select_tx;
225 	rs->rounds_select_intr = mtcp->runstat.rounds_select_intr - mtcp->p_runstat.rounds_select_intr;
226 	rs->rounds_twcheck = mtcp->runstat.rounds_twcheck - mtcp->p_runstat.rounds_twcheck;
227 	mtcp->p_runstat = mtcp->runstat;
228 #if NETSTAT_PERTHREAD
229 	fprintf(stderr, "[CPU%2d] Rounds: %4lluK, "
230 			"rx: %3lluK (try: %4lluK), tx: %3lluK (try: %4lluK), "
231 			"ps_select: %4llu (rx: %4llu, tx: %4llu, intr: %3llu)\n",
232 			mtcp->ctx->cpu, rs->rounds / ROUND_DIV,
233 			rs->rounds_rx / ROUND_DIV, rs->rounds_rx_try / ROUND_DIV,
234 			rs->rounds_tx / ROUND_DIV, rs->rounds_tx_try / ROUND_DIV,
235 			rs->rounds_select,
236 			rs->rounds_select_rx, rs->rounds_select_tx, rs->rounds_select_intr);
237 #endif
238 }
239 #endif /* ROUND_STAT */
240 /*----------------------------------------------------------------------------*/
241 #if TIME_STAT
242 static inline void
243 PrintThreadRoundTime(mtcp_manager_t mtcp)
244 {
245 	fprintf(stderr, "[CPU%2d] Time: (avg, max) "
246 			"round: (%4luus, %4luus), processing: (%4luus, %4luus), "
247 			"tcheck: (%4luus, %4luus), epoll: (%4luus, %4luus), "
248 			"handle: (%4luus, %4luus), xmit: (%4luus, %4luus), "
249 			"select: (%4luus, %4luus)\n", mtcp->ctx->cpu,
250 			GetAverageStat(&mtcp->rtstat.round), mtcp->rtstat.round.max,
251 			GetAverageStat(&mtcp->rtstat.processing), mtcp->rtstat.processing.max,
252 			GetAverageStat(&mtcp->rtstat.tcheck), mtcp->rtstat.tcheck.max,
253 			GetAverageStat(&mtcp->rtstat.epoll), mtcp->rtstat.epoll.max,
254 			GetAverageStat(&mtcp->rtstat.handle), mtcp->rtstat.handle.max,
255 			GetAverageStat(&mtcp->rtstat.xmit), mtcp->rtstat.xmit.max,
256 			GetAverageStat(&mtcp->rtstat.select), mtcp->rtstat.select.max);
257 
258 	InitStatCounter(&mtcp->rtstat.round);
259 	InitStatCounter(&mtcp->rtstat.processing);
260 	InitStatCounter(&mtcp->rtstat.tcheck);
261 	InitStatCounter(&mtcp->rtstat.epoll);
262 	InitStatCounter(&mtcp->rtstat.handle);
263 	InitStatCounter(&mtcp->rtstat.xmit);
264 	InitStatCounter(&mtcp->rtstat.select);
265 }
266 #endif
267 #endif /* NETSTAT */
268 /*----------------------------------------------------------------------------*/
269 #if EVENT_STAT
270 static inline void
271 PrintEventStat(int core, struct mtcp_epoll_stat *stat)
272 {
273 	fprintf(stderr, "[CPU%2d] calls: %lu, waits: %lu, wakes: %lu, "
274 			"issued: %lu, registered: %lu, invalidated: %lu, handled: %lu\n",
275 			core, stat->calls, stat->waits, stat->wakes,
276 			stat->issued, stat->registered, stat->invalidated, stat->handled);
277 	memset(stat, 0, sizeof(struct mtcp_epoll_stat));
278 }
279 #endif /* EVENT_STAT */
280 /*----------------------------------------------------------------------------*/
281 #ifdef NETSTAT
282 static inline void
283 PrintNetworkStats(mtcp_manager_t mtcp, uint32_t cur_ts)
284 {
285 #define TIMEOUT 1
286 	int i;
287 	struct net_stat ns;
288 	bool stat_print = false;
289 #if ROUND_STAT
290 	struct run_stat rs;
291 #endif /* ROUND_STAT */
292 #ifdef NETSTAT_TOTAL
293 	static double peak_total_rx_gbps = 0;
294 	static double peak_total_tx_gbps = 0;
295 	static double avg_total_rx_gbps = 0;
296 	static double avg_total_tx_gbps = 0;
297 
298 	double total_rx_gbps = 0, total_tx_gbps = 0;
299 	int j;
300 	uint32_t gflow_cnt = 0;
301 	struct net_stat g_nstat;
302 #if ROUND_STAT
303 	struct run_stat g_runstat;
304 #endif /* ROUND_STAT */
305 #endif /* NETSTAT_TOTAL */
306 
307 	if (TS_TO_MSEC(cur_ts - mtcp->p_nstat_ts) < SEC_TO_MSEC(TIMEOUT)) {
308 		return;
309 	}
310 
311 	mtcp->p_nstat_ts = cur_ts;
312 	gflow_cnt = 0;
313 	memset(&g_nstat, 0, sizeof(struct net_stat));
314 	for (i = 0; i < g_config.mos->num_cores; i++) {
315 		if (running[i]) {
316 			PrintThreadNetworkStats(g_mtcp[i], &ns);
317 #if NETSTAT_TOTAL
318 			gflow_cnt += g_mtcp[i]->flow_cnt;
319 			for (j = 0; j < g_config.mos->netdev_table->num; j++) {
320 				g_nstat.rx_packets[j] += ns.rx_packets[j];
321 				g_nstat.rx_errors[j] += ns.rx_errors[j];
322 				g_nstat.rx_bytes[j] += ns.rx_bytes[j];
323 				g_nstat.tx_packets[j] += ns.tx_packets[j];
324 				g_nstat.tx_drops[j] += ns.tx_drops[j];
325 				g_nstat.tx_bytes[j] += ns.tx_bytes[j];
326 			}
327 #endif
328 		}
329 	}
330 #if NETSTAT_TOTAL
331 	for (i = 0; i < g_config.mos->netdev_table->num; i++) {
332 		if (g_config.mos->netdev_table->ent[i]->stat_print) {
333 			fprintf(stderr, "[ ALL ] %s, "
334 			"RX: %7llu(pps) (err: %5llu), %5.2lf(Gbps), "
335 			"TX: %7llu(pps), %5.2lf(Gbps)\n",
336 				g_config.mos->netdev_table->ent[i]->dev_name,
337 				(long long unsigned)g_nstat.rx_packets[i],
338 				(long long unsigned)g_nstat.rx_errors[i],
339 				GBPS(g_nstat.rx_bytes[i]),
340 				(long long unsigned)g_nstat.tx_packets[i],
341 				GBPS(g_nstat.tx_bytes[i]));
342 			total_rx_gbps += GBPS(g_nstat.rx_bytes[i]);
343 			total_tx_gbps += GBPS(g_nstat.tx_bytes[i]);
344 			stat_print = true;
345 		}
346 	}
347 	if (stat_print) {
348 		fprintf(stderr, "[ ALL ] flows: %6u\n", gflow_cnt);
349 		if (avg_total_rx_gbps == 0)
350 			avg_total_rx_gbps = total_rx_gbps;
351 		else
352 			avg_total_rx_gbps = avg_total_rx_gbps * 0.6 + total_rx_gbps * 0.4;
353 
354 		if (avg_total_tx_gbps == 0)
355 			avg_total_tx_gbps = total_tx_gbps;
356 		else
357 			avg_total_tx_gbps = avg_total_tx_gbps * 0.6 + total_tx_gbps * 0.4;
358 
359 		if (peak_total_rx_gbps < total_rx_gbps)
360 			peak_total_rx_gbps = total_rx_gbps;
361 		if (peak_total_tx_gbps < total_tx_gbps)
362 			peak_total_tx_gbps = total_tx_gbps;
363 
364 		fprintf(stderr, "[ PEAK ] RX: %5.2lf(Gbps), TX: %5.2lf(Gbps)\n"
365 						"[ RECENT AVG ] RX: %5.2lf(Gbps), TX: %5.2lf(Gbps)\n",
366 				peak_total_rx_gbps, peak_total_tx_gbps,
367 				avg_total_rx_gbps, avg_total_tx_gbps);
368 	}
369 #endif
370 
371 #if ROUND_STAT
372 	memset(&g_runstat, 0, sizeof(struct run_stat));
373 	for (i = 0; i < g_config.mos->num_cores; i++) {
374 		if (running[i]) {
375 			PrintThreadRoundStats(g_mtcp[i], &rs);
376 #if DBGMSG
377 			g_runstat.rounds += rs.rounds;
378 			g_runstat.rounds_rx += rs.rounds_rx;
379 			g_runstat.rounds_rx_try += rs.rounds_rx_try;
380 			g_runstat.rounds_tx += rs.rounds_tx;
381 			g_runstat.rounds_tx_try += rs.rounds_tx_try;
382 			g_runstat.rounds_select += rs.rounds_select;
383 			g_runstat.rounds_select_rx += rs.rounds_select_rx;
384 			g_runstat.rounds_select_tx += rs.rounds_select_tx;
385 #endif
386 		}
387 	}
388 
389 	TRACE_DBG("[ ALL ] Rounds: %4ldK, "
390 		  "rx: %3ldK (try: %4ldK), tx: %3ldK (try: %4ldK), "
391 		  "ps_select: %4ld (rx: %4ld, tx: %4ld)\n",
392 		  g_runstat.rounds / 1000, g_runstat.rounds_rx / 1000,
393 		  g_runstat.rounds_rx_try / 1000, g_runstat.rounds_tx / 1000,
394 		  g_runstat.rounds_tx_try / 1000, g_runstat.rounds_select,
395 		  g_runstat.rounds_select_rx, g_runstat.rounds_select_tx);
396 #endif /* ROUND_STAT */
397 
398 #if TIME_STAT
399 	for (i = 0; i < g_config.mos->num_cores; i++) {
400 		if (running[i]) {
401 			PrintThreadRoundTime(g_mtcp[i]);
402 		}
403 	}
404 #endif
405 
406 #if EVENT_STAT
407 	for (i = 0; i < g_config.mos->num_cores; i++) {
408 		if (running[i] && g_mtcp[i]->ep) {
409 			PrintEventStat(i, &g_mtcp[i]->ep->stat);
410 		}
411 	}
412 #endif
413 
414 	fflush(stderr);
415 }
416 #endif /* NETSTAT */
417 /*----------------------------------------------------------------------------*/
418 static inline void
419 FlushMonitorReadEvents(mtcp_manager_t mtcp)
420 {
421 	struct event_queue *mtcpq;
422 	struct tcp_stream *cur_stream;
423 	struct mon_listener *walk;
424 
425 	/* check if monitor sockets should be passed data */
426 	TAILQ_FOREACH(walk, &mtcp->monitors, link) {
427 		if (walk->socket->socktype != MOS_SOCK_MONITOR_STREAM ||
428 			!(mtcpq = walk->eq))
429 			continue;
430 
431 		while (mtcpq->num_events > 0) {
432 			cur_stream =
433 				(struct tcp_stream *)mtcpq->events[mtcpq->start++].ev.data.ptr;
434 			/* only read events */
435 			if (cur_stream != NULL &&
436 					(cur_stream->socket->events | MOS_EPOLLIN)) {
437 				if (cur_stream->rcvvar != NULL &&
438 						cur_stream->rcvvar->rcvbuf != NULL) {
439 					/* no need to pass pkt context */
440 					struct socket_map *walk;
441 					SOCKQ_FOREACH_START(walk, &cur_stream->msocks) {
442 						HandleCallback(mtcp, MOS_NULL, walk,
443 							       cur_stream->side, NULL,
444 							       MOS_ON_CONN_NEW_DATA);
445 					} SOCKQ_FOREACH_END;
446 				}
447 				/* reset the actions now */
448 				cur_stream->actions = 0;
449 			}
450 			if (mtcpq->start >= mtcpq->size)
451 				mtcpq->start = 0;
452 			mtcpq->num_events--;
453 		}
454 	}
455 }
456 /*----------------------------------------------------------------------------*/
457 static inline void
458 FlushBufferedReadEvents(mtcp_manager_t mtcp)
459 {
460 	int i;
461 	int offset;
462 	struct event_queue *mtcpq;
463 	struct tcp_stream *cur_stream;
464 
465 	if (mtcp->ep == NULL) {
466 		TRACE_EPOLL("No epoll socket has been registered yet!\n");
467 		return;
468 	} else {
469 		/* case when mtcpq exists */
470 		mtcpq = mtcp->ep->mtcp_queue;
471 		offset = mtcpq->start;
472 	}
473 
474 	/* we will use queued-up epoll read-in events
475 	 * to trigger buffered read monitor events */
476 	for (i = 0; i < mtcpq->num_events; i++) {
477 		cur_stream = mtcp->smap[mtcpq->events[offset++].sockid].stream;
478 		/* only read events */
479 		/* Raise new data callback event */
480 		if (cur_stream != NULL &&
481 				(cur_stream->socket->events | MOS_EPOLLIN)) {
482 			if (cur_stream->rcvvar != NULL &&
483 					cur_stream->rcvvar->rcvbuf != NULL) {
484 				/* no need to pass pkt context */
485 				struct socket_map *walk;
486 				SOCKQ_FOREACH_START(walk, &cur_stream->msocks) {
487 					HandleCallback(mtcp, MOS_NULL, walk, cur_stream->side,
488 							NULL, MOS_ON_CONN_NEW_DATA);
489 				} SOCKQ_FOREACH_END;
490 			}
491 		}
492 		if (offset >= mtcpq->size)
493 			offset = 0;
494 	}
495 }
496 /*----------------------------------------------------------------------------*/
497 static inline void
498 FlushEpollEvents(mtcp_manager_t mtcp, uint32_t cur_ts)
499 {
500 	struct mtcp_epoll *ep = mtcp->ep;
501 	struct event_queue *usrq = ep->usr_queue;
502 	struct event_queue *mtcpq = ep->mtcp_queue;
503 
504 	pthread_mutex_lock(&ep->epoll_lock);
505 	if (ep->mtcp_queue->num_events > 0) {
506 		/* while mtcp_queue have events */
507 		/* and usr_queue is not full */
508 		while (mtcpq->num_events > 0 && usrq->num_events < usrq->size) {
509 			/* copy the event from mtcp_queue to usr_queue */
510 			usrq->events[usrq->end++] = mtcpq->events[mtcpq->start++];
511 
512 			if (usrq->end >= usrq->size)
513 				usrq->end = 0;
514 			usrq->num_events++;
515 
516 			if (mtcpq->start >= mtcpq->size)
517 				mtcpq->start = 0;
518 			mtcpq->num_events--;
519 		}
520 	}
521 
522 	/* if there are pending events, wake up user */
523 	if (ep->waiting && (ep->usr_queue->num_events > 0 ||
524 				ep->usr_shadow_queue->num_events > 0)) {
525 		STAT_COUNT(mtcp->runstat.rounds_epoll);
526 		TRACE_EPOLL("Broadcasting events. num: %d, cur_ts: %u, prev_ts: %u\n",
527 				ep->usr_queue->num_events, cur_ts, mtcp->ts_last_event);
528 		mtcp->ts_last_event = cur_ts;
529 		ep->stat.wakes++;
530 		pthread_cond_signal(&ep->epoll_cond);
531 	}
532 	pthread_mutex_unlock(&ep->epoll_lock);
533 }
534 /*----------------------------------------------------------------------------*/
535 static inline void
536 HandleApplicationCalls(mtcp_manager_t mtcp, uint32_t cur_ts)
537 {
538 	tcp_stream *stream;
539 	int cnt, max_cnt;
540 	int handled, delayed;
541 	int control, send, ack;
542 
543 	/* connect handling */
544 	while ((stream = StreamDequeue(mtcp->connectq))) {
545 		if (stream->state != TCP_ST_SYN_SENT) {
546 			TRACE_INFO("Got a connection request from app with state: %s",
547 				   TCPStateToString(stream));
548 			exit(EXIT_FAILURE);
549 		} else {
550 			stream->cb_events |= MOS_ON_CONN_START |
551 				MOS_ON_TCP_STATE_CHANGE;
552 			/* if monitor is on... */
553 			if (stream->pair_stream != NULL)
554 				stream->pair_stream->cb_events |=
555 					MOS_ON_CONN_START;
556 		}
557 		AddtoControlList(mtcp, stream, cur_ts);
558 	}
559 
560 	/* send queue handling */
561 	while ((stream = StreamDequeue(mtcp->sendq))) {
562 		stream->sndvar->on_sendq = FALSE;
563 		AddtoSendList(mtcp, stream);
564 	}
565 
566 	/* ack queue handling */
567 	while ((stream = StreamDequeue(mtcp->ackq))) {
568 		stream->sndvar->on_ackq = FALSE;
569 		EnqueueACK(mtcp, stream, cur_ts, ACK_OPT_AGGREGATE);
570 	}
571 
572 	/* close handling */
573 	handled = delayed = 0;
574 	control = send = ack = 0;
575 	while ((stream = StreamDequeue(mtcp->closeq))) {
576 		struct tcp_send_vars *sndvar = stream->sndvar;
577 		sndvar->on_closeq = FALSE;
578 
579 		if (sndvar->sndbuf) {
580 			sndvar->fss = sndvar->sndbuf->head_seq + sndvar->sndbuf->len;
581 		} else {
582 			sndvar->fss = stream->snd_nxt;
583 		}
584 
585 		if (g_config.mos->tcp_timeout > 0)
586 			RemoveFromTimeoutList(mtcp, stream);
587 
588 		if (stream->have_reset) {
589 			handled++;
590 			if (stream->state != TCP_ST_CLOSED_RSVD) {
591 				stream->close_reason = TCP_RESET;
592 				stream->state = TCP_ST_CLOSED_RSVD;
593 				stream->cb_events |= MOS_ON_TCP_STATE_CHANGE;
594 				TRACE_STATE("Stream %d: TCP_ST_CLOSED_RSVD\n", stream->id);
595 				DestroyTCPStream(mtcp, stream);
596 			} else {
597 				TRACE_ERROR("Stream already closed.\n");
598 			}
599 
600 		} else if (sndvar->on_control_list) {
601 			sndvar->on_closeq_int = TRUE;
602 			StreamInternalEnqueue(mtcp->closeq_int, stream);
603 			delayed++;
604 			if (sndvar->on_control_list)
605 				control++;
606 			if (sndvar->on_send_list)
607 				send++;
608 			if (sndvar->on_ack_list)
609 				ack++;
610 
611 		} else if (sndvar->on_send_list || sndvar->on_ack_list) {
612 			handled++;
613 			if (stream->state == TCP_ST_ESTABLISHED) {
614 				stream->state = TCP_ST_FIN_WAIT_1;
615 				stream->cb_events |= MOS_ON_TCP_STATE_CHANGE;
616 				TRACE_STATE("Stream %d: TCP_ST_FIN_WAIT_1\n", stream->id);
617 
618 			} else if (stream->state == TCP_ST_CLOSE_WAIT) {
619 				stream->state = TCP_ST_LAST_ACK;
620 				stream->cb_events |= MOS_ON_TCP_STATE_CHANGE;
621 				TRACE_STATE("Stream %d: TCP_ST_LAST_ACK\n", stream->id);
622 			}
623 			stream->control_list_waiting = TRUE;
624 
625 		} else if (stream->state != TCP_ST_CLOSED_RSVD) {
626 			handled++;
627 			if (stream->state == TCP_ST_ESTABLISHED) {
628 				stream->state = TCP_ST_FIN_WAIT_1;
629 				stream->cb_events |= MOS_ON_TCP_STATE_CHANGE;
630 				TRACE_STATE("Stream %d: TCP_ST_FIN_WAIT_1\n", stream->id);
631 
632 			} else if (stream->state == TCP_ST_CLOSE_WAIT) {
633 				stream->state = TCP_ST_LAST_ACK;
634 				stream->cb_events |= MOS_ON_TCP_STATE_CHANGE;
635 				TRACE_STATE("Stream %d: TCP_ST_LAST_ACK\n", stream->id);
636 			}
637 			//sndvar->rto = TCP_FIN_RTO;
638 			//UpdateRetransmissionTimer(mtcp, stream, mtcp->cur_ts);
639 			AddtoControlList(mtcp, stream, cur_ts);
640 		} else {
641 			TRACE_ERROR("Already closed connection!\n");
642 		}
643 	}
644 	TRACE_ROUND("Handling close connections. cnt: %d\n", cnt);
645 
646 	cnt = 0;
647 	max_cnt = mtcp->closeq_int->count;
648 	while (cnt++ < max_cnt) {
649 		stream = StreamInternalDequeue(mtcp->closeq_int);
650 
651 		if (stream->sndvar->on_control_list) {
652 			StreamInternalEnqueue(mtcp->closeq_int, stream);
653 
654 		} else if (stream->state != TCP_ST_CLOSED_RSVD) {
655 			handled++;
656 			stream->sndvar->on_closeq_int = FALSE;
657 			if (stream->state == TCP_ST_ESTABLISHED) {
658 				stream->state = TCP_ST_FIN_WAIT_1;
659 				stream->cb_events |= MOS_ON_TCP_STATE_CHANGE;
660 				TRACE_STATE("Stream %d: TCP_ST_FIN_WAIT_1\n", stream->id);
661 
662 			} else if (stream->state == TCP_ST_CLOSE_WAIT) {
663 				stream->state = TCP_ST_LAST_ACK;
664 				stream->cb_events |= MOS_ON_TCP_STATE_CHANGE;
665 				TRACE_STATE("Stream %d: TCP_ST_LAST_ACK\n", stream->id);
666 			}
667 			AddtoControlList(mtcp, stream, cur_ts);
668 		} else {
669 			stream->sndvar->on_closeq_int = FALSE;
670 			TRACE_ERROR("Already closed connection!\n");
671 		}
672 	}
673 
674 	/* reset handling */
675 	while ((stream = StreamDequeue(mtcp->resetq))) {
676 		stream->sndvar->on_resetq = FALSE;
677 
678 		if (g_config.mos->tcp_timeout > 0)
679 			RemoveFromTimeoutList(mtcp, stream);
680 
681 		if (stream->have_reset) {
682 			if (stream->state != TCP_ST_CLOSED_RSVD) {
683 				stream->close_reason = TCP_RESET;
684 				stream->state = TCP_ST_CLOSED_RSVD;
685 				stream->cb_events |= MOS_ON_TCP_STATE_CHANGE;
686 				TRACE_STATE("Stream %d: TCP_ST_CLOSED_RSVD\n", stream->id);
687 				DestroyTCPStream(mtcp, stream);
688 			} else {
689 				TRACE_ERROR("Stream already closed.\n");
690 			}
691 
692 		} else if (stream->sndvar->on_control_list ||
693 				stream->sndvar->on_send_list || stream->sndvar->on_ack_list) {
694 			/* wait until all the queues are flushed */
695 			stream->sndvar->on_resetq_int = TRUE;
696 			StreamInternalEnqueue(mtcp->resetq_int, stream);
697 
698 		} else {
699 			if (stream->state != TCP_ST_CLOSED_RSVD) {
700 				stream->close_reason = TCP_ACTIVE_CLOSE;
701 				stream->state = TCP_ST_CLOSED_RSVD;
702 				stream->cb_events |= MOS_ON_TCP_STATE_CHANGE;
703 				TRACE_STATE("Stream %d: TCP_ST_CLOSED_RSVD\n", stream->id);
704 				AddtoControlList(mtcp, stream, cur_ts);
705 			} else {
706 				TRACE_ERROR("Stream already closed.\n");
707 			}
708 		}
709 	}
710 	TRACE_ROUND("Handling reset connections. cnt: %d\n", cnt);
711 
712 	cnt = 0;
713 	max_cnt = mtcp->resetq_int->count;
714 	while (cnt++ < max_cnt) {
715 		stream = StreamInternalDequeue(mtcp->resetq_int);
716 
717 		if (stream->sndvar->on_control_list ||
718 				stream->sndvar->on_send_list || stream->sndvar->on_ack_list) {
719 			/* wait until all the queues are flushed */
720 			StreamInternalEnqueue(mtcp->resetq_int, stream);
721 
722 		} else {
723 			stream->sndvar->on_resetq_int = FALSE;
724 
725 			if (stream->state != TCP_ST_CLOSED_RSVD) {
726 				stream->close_reason = TCP_ACTIVE_CLOSE;
727 				stream->state = TCP_ST_CLOSED_RSVD;
728 				stream->cb_events |= MOS_ON_TCP_STATE_CHANGE;
729 				TRACE_STATE("Stream %d: TCP_ST_CLOSED_RSVD\n", stream->id);
730 				AddtoControlList(mtcp, stream, cur_ts);
731 			} else {
732 				TRACE_ERROR("Stream already closed.\n");
733 			}
734 		}
735 	}
736 
737 	/* destroy streams in destroyq */
738 	while ((stream = StreamDequeue(mtcp->destroyq))) {
739 		DestroyTCPStream(mtcp, stream);
740 	}
741 
742 	mtcp->wakeup_flag = FALSE;
743 }
744 /*----------------------------------------------------------------------------*/
745 static inline void
746 WritePacketsToChunks(mtcp_manager_t mtcp, uint32_t cur_ts)
747 {
748 	int thresh = g_config.mos->max_concurrency;
749 	int i;
750 
751 	/* Set the threshold to g_config.mos->max_concurrency to send ACK immediately */
752 	/* Otherwise, set to appropriate value (e.g. thresh) */
753 	assert(mtcp->g_sender != NULL);
754 	if (mtcp->g_sender->control_list_cnt)
755 		WriteTCPControlList(mtcp, mtcp->g_sender, cur_ts, thresh);
756 	if (mtcp->g_sender->ack_list_cnt)
757 		WriteTCPACKList(mtcp, mtcp->g_sender, cur_ts, thresh);
758 	if (mtcp->g_sender->send_list_cnt)
759 		WriteTCPDataList(mtcp, mtcp->g_sender, cur_ts, thresh);
760 
761 	for (i = 0; i < g_config.mos->netdev_table->num; i++) {
762 		assert(mtcp->n_sender[i] != NULL);
763 		if (mtcp->n_sender[i]->control_list_cnt)
764 			WriteTCPControlList(mtcp, mtcp->n_sender[i], cur_ts, thresh);
765 		if (mtcp->n_sender[i]->ack_list_cnt)
766 			WriteTCPACKList(mtcp, mtcp->n_sender[i], cur_ts, thresh);
767 		if (mtcp->n_sender[i]->send_list_cnt)
768 			WriteTCPDataList(mtcp, mtcp->n_sender[i], cur_ts, thresh);
769 	}
770 }
771 /*----------------------------------------------------------------------------*/
772 #if TESTING
773 static int
774 DestroyRemainingFlows(mtcp_manager_t mtcp)
775 {
776 	struct hashtable *ht = mtcp->tcp_flow_table;
777 	tcp_stream *walk;
778 	int cnt, i;
779 
780 	cnt = 0;
781 
782 	thread_printf(mtcp, mtcp->log_fp,
783 			"CPU %d: Flushing remaining flows.\n", mtcp->ctx->cpu);
784 
785 	for (i = 0; i < NUM_BINS; i++) {
786 		TAILQ_FOREACH(walk, &ht->ht_table[i], rcvvar->he_link) {
787 			thread_printf(mtcp, mtcp->log_fp,
788 					"CPU %d: Destroying stream %d\n", mtcp->ctx->cpu, walk->id);
789 			DumpStream(mtcp, walk);
790 			DestroyTCPStream(mtcp, walk);
791 			cnt++;
792 		}
793 	}
794 
795 	return cnt;
796 }
797 #endif
798 /*----------------------------------------------------------------------------*/
799 static void
800 InterruptApplication(mtcp_manager_t mtcp)
801 {
802 	/* interrupt if the mtcp_epoll_wait() is waiting */
803 	if (mtcp->ep) {
804 		pthread_mutex_lock(&mtcp->ep->epoll_lock);
805 		if (mtcp->ep->waiting) {
806 			pthread_cond_signal(&mtcp->ep->epoll_cond);
807 		}
808 		pthread_mutex_unlock(&mtcp->ep->epoll_lock);
809 	}
810 	/* interrupt if the accept() is waiting */
811 	if (mtcp->listener) {
812 		if (mtcp->listener->socket) {
813 			pthread_mutex_lock(&mtcp->listener->accept_lock);
814 			if (!(mtcp->listener->socket->opts & MTCP_NONBLOCK)) {
815 				pthread_cond_signal(&mtcp->listener->accept_cond);
816 			}
817 			pthread_mutex_unlock(&mtcp->listener->accept_lock);
818 		}
819 	}
820 }
821 /*----------------------------------------------------------------------------*/
822 void
823 RunPassiveLoop(mtcp_manager_t mtcp)
824 {
825 	sem_wait(&g_done_sem[mtcp->ctx->cpu]);
826 	sem_destroy(&g_done_sem[mtcp->ctx->cpu]);
827 	return;
828 }
829 /*----------------------------------------------------------------------------*/
830 static void
831 RunMainLoop(struct mtcp_thread_context *ctx)
832 {
833 	mtcp_manager_t mtcp = ctx->mtcp_manager;
834 	int i;
835 	int recv_cnt;
836 
837 #if E_PSIO
838 	int rx_inf, tx_inf;
839 #endif
840 
841 	struct timeval cur_ts = {0};
842 	uint32_t ts, ts_prev;
843 
844 #if TIME_STAT
845 	struct timeval prev_ts, processing_ts, tcheck_ts,
846 				   epoll_ts, handle_ts, xmit_ts, select_ts;
847 #endif
848 	int thresh;
849 
850 	gettimeofday(&cur_ts, NULL);
851 
852 #if E_PSIO
853 	/* do nothing */
854 #else
855 #if !USE_CHUNK_BUF
856 	/* create packet write chunk */
857 	InitWriteChunks(handle, ctx->w_chunk);
858 	for (i = 0; i < ETH_NUM; i++) {
859 		ctx->w_chunk[i].cnt = 0;
860 		ctx->w_off[i] = 0;
861 		ctx->w_cur_idx[i] = 0;
862 	}
863 #endif
864 #endif
865 
866 	TRACE_DBG("CPU %d: mtcp thread running.\n", ctx->cpu);
867 
868 #if TIME_STAT
869 	prev_ts = cur_ts;
870 	InitStatCounter(&mtcp->rtstat.round);
871 	InitStatCounter(&mtcp->rtstat.processing);
872 	InitStatCounter(&mtcp->rtstat.tcheck);
873 	InitStatCounter(&mtcp->rtstat.epoll);
874 	InitStatCounter(&mtcp->rtstat.handle);
875 	InitStatCounter(&mtcp->rtstat.xmit);
876 	InitStatCounter(&mtcp->rtstat.select);
877 #endif
878 
879 	ts = ts_prev = 0;
880 	while ((!ctx->done || mtcp->flow_cnt) && !ctx->exit) {
881 
882 		STAT_COUNT(mtcp->runstat.rounds);
883 		recv_cnt = 0;
884 
885 #if E_PSIO
886 #if 0
887 		event.timeout = PS_SELECT_TIMEOUT;
888 		NID_ZERO(event.rx_nids);
889 		NID_ZERO(event.tx_nids);
890 		NID_ZERO(rx_avail);
891 		//NID_ZERO(tx_avail);
892 #endif
893 		gettimeofday(&cur_ts, NULL);
894 #if TIME_STAT
895 		/* measure the inter-round delay */
896 		UpdateStatCounter(&mtcp->rtstat.round, TimeDiffUs(&cur_ts, &prev_ts));
897 		prev_ts = cur_ts;
898 #endif
899 
900 		ts = TIMEVAL_TO_TS(&cur_ts);
901 		mtcp->cur_ts = ts;
902 
903 		for (rx_inf = 0; rx_inf < g_config.mos->netdev_table->num; rx_inf++) {
904 
905 			recv_cnt = mtcp->iom->recv_pkts(ctx, rx_inf);
906 			STAT_COUNT(mtcp->runstat.rounds_rx_try);
907 
908 			for (i = 0; i < recv_cnt; i++) {
909 				uint16_t len;
910 				uint8_t *pktbuf;
911 				pktbuf = mtcp->iom->get_rptr(mtcp->ctx, rx_inf, i, &len);
912 				ProcessPacket(mtcp, rx_inf, i, ts, pktbuf, len);
913 			}
914 
915 #ifdef ENABLE_DPDKR
916 			mtcp->iom->send_pkts(ctx, rx_inf);
917 			continue;
918 #endif
919 		}
920 		STAT_COUNT(mtcp->runstat.rounds_rx);
921 
922 #else /* E_PSIO */
923 		gettimeofday(&cur_ts, NULL);
924 		ts = TIMEVAL_TO_TS(&cur_ts);
925 		mtcp->cur_ts = ts;
926 		/*
927 		 * Read packets into a chunk from NIC
928 		 * chk_w_idx : next chunk index to write packets from NIC
929 		 */
930 
931 		STAT_COUNT(mtcp->runstat.rounds_rx_try);
932 		chunk.cnt = PS_CHUNK_SIZE;
933 		recv_cnt = ps_recv_chunk(handle, &chunk);
934 		if (recv_cnt < 0) {
935 			if (errno != EAGAIN && errno != EINTR) {
936 				perror("ps_recv_chunk");
937 				assert(0);
938 			}
939 		}
940 
941 		/*
942 		 * Handling Packets
943 		 * chk_r_idx : next chunk index to read and handle
944 		*/
945 		for (i = 0; i < recv_cnt; i++) {
946 			ProcessPacket(mtcp, chunk.queue.ifindex, ts,
947 					(u_char *)(chunk.buf + chunk.info[i].offset),
948 					chunk.info[i].len);
949 		}
950 
951 		if (recv_cnt > 0)
952 			STAT_COUNT(mtcp->runstat.rounds_rx);
953 #endif /* E_PSIO */
954 #if TIME_STAT
955 		gettimeofday(&processing_ts, NULL);
956 		UpdateStatCounter(&mtcp->rtstat.processing,
957 				TimeDiffUs(&processing_ts, &cur_ts));
958 #endif /* TIME_STAT */
959 
960 		/* Handle user defined timeout */
961 		struct timer *walk, *tmp;
962 		for (walk = TAILQ_FIRST(&mtcp->timer_list); walk != NULL; walk = tmp) {
963 			tmp = TAILQ_NEXT(walk, timer_link);
964 			if (TIMEVAL_LT(&cur_ts, &walk->exp))
965 				break;
966 
967 			struct mtcp_context mctx = {.cpu = ctx->cpu};
968 			walk->cb(&mctx, walk->id, 0, 0 /* FIXME */, NULL);
969 			DelTimer(mtcp, walk);
970 		}
971 
972 		/* interaction with application */
973 		if (mtcp->flow_cnt > 0) {
974 
975 			/* check retransmission timeout and timewait expire */
976 #if 0
977 			thresh = (int)mtcp->flow_cnt / (TS_TO_USEC(PER_STREAM_TCHECK));
978 			assert(thresh >= 0);
979 			if (thresh == 0)
980 				thresh = 1;
981 			if (recv_cnt > 0 && thresh > recv_cnt)
982 				thresh = recv_cnt;
983 #else
984 			thresh = g_config.mos->max_concurrency;
985 #endif
986 
987 			/* Eunyoung, you may fix this later
988 			 * if there is no rcv packet, we will send as much as possible
989 			 */
990 			if (thresh == -1)
991 				thresh = g_config.mos->max_concurrency;
992 
993 			CheckRtmTimeout(mtcp, ts, thresh);
994 			CheckTimewaitExpire(mtcp, ts, thresh);
995 
996 			if (g_config.mos->tcp_timeout > 0 && ts != ts_prev) {
997 				CheckConnectionTimeout(mtcp, ts, thresh);
998 			}
999 
1000 #if TIME_STAT
1001 		}
1002 		gettimeofday(&tcheck_ts, NULL);
1003 		UpdateStatCounter(&mtcp->rtstat.tcheck,
1004 				TimeDiffUs(&tcheck_ts, &processing_ts));
1005 
1006 		if (mtcp->flow_cnt > 0) {
1007 #endif /* TIME_STAT */
1008 
1009 		}
1010 
1011 		/*
1012 		 * before flushing epoll events, call monitor events for
1013 		 * all registered `read` events
1014 		 */
1015 		if (mtcp->num_msp > 0)
1016 			/* call this when only a standalone monitor is running */
1017 			FlushMonitorReadEvents(mtcp);
1018 
1019 		/* if epoll is in use, flush all the queued events */
1020 		if (mtcp->ep) {
1021 			FlushBufferedReadEvents(mtcp);
1022 			FlushEpollEvents(mtcp, ts);
1023 		}
1024 #if TIME_STAT
1025 		gettimeofday(&epoll_ts, NULL);
1026 		UpdateStatCounter(&mtcp->rtstat.epoll,
1027 				TimeDiffUs(&epoll_ts, &tcheck_ts));
1028 #endif /* TIME_STAT */
1029 
1030 		if (end_app_exists && mtcp->flow_cnt > 0) {
1031 			/* handle stream queues  */
1032 			HandleApplicationCalls(mtcp, ts);
1033 		}
1034 
1035 #ifdef ENABLE_DPDKR
1036 		continue;
1037 #endif
1038 
1039 #if TIME_STAT
1040 		gettimeofday(&handle_ts, NULL);
1041 		UpdateStatCounter(&mtcp->rtstat.handle,
1042 				TimeDiffUs(&handle_ts, &epoll_ts));
1043 #endif /* TIME_STAT */
1044 
1045 		WritePacketsToChunks(mtcp, ts);
1046 
1047 		/* send packets from write buffer */
1048 #if E_PSIO
1049 		/* With E_PSIO, send until tx is available */
1050 		int num_dev = g_config.mos->netdev_table->num;
1051 		if (likely(mtcp->iom->send_pkts != NULL))
1052 			for (tx_inf = 0; tx_inf < num_dev; tx_inf++) {
1053 				mtcp->iom->send_pkts(ctx, tx_inf);
1054 			}
1055 
1056 #else /* E_PSIO */
1057 		/* Without E_PSIO, try send chunks immediately */
1058 		for (i = 0; i < g_config.mos->netdev_table->num; i++) {
1059 #if USE_CHUNK_BUF
1060 			/* in the case of using ps_send_chunk_buf() without E_PSIO */
1061 			ret = FlushSendChunkBuf(mtcp, i);
1062 #else
1063 			/* if not using ps_send_chunk_buf() */
1064 			ret = FlushWriteBuffer(ctx, i);
1065 #endif
1066 			if (ret < 0) {
1067 				TRACE_ERROR("Failed to send chunks.\n");
1068 			} else if (ret > 0) {
1069 				STAT_COUNT(mtcp->runstat.rounds_tx);
1070 			}
1071 		}
1072 #endif /* E_PSIO */
1073 
1074 #if TIME_STAT
1075 		gettimeofday(&xmit_ts, NULL);
1076 		UpdateStatCounter(&mtcp->rtstat.xmit,
1077 				TimeDiffUs(&xmit_ts, &handle_ts));
1078 #endif /* TIME_STAT */
1079 
1080 		if (ts != ts_prev) {
1081 			ts_prev = ts;
1082 #ifdef NETSTAT
1083 			if (ctx->cpu == printer) {
1084 #ifdef RUN_ARP
1085 				ARPTimer(mtcp, ts);
1086 #endif
1087 #ifdef NETSTAT
1088 				PrintNetworkStats(mtcp, ts);
1089 #endif
1090 			}
1091 #endif /* NETSTAT */
1092 		}
1093 
1094 		if (mtcp->iom->select)
1095 			mtcp->iom->select(ctx);
1096 
1097 		if (ctx->interrupt) {
1098 			InterruptApplication(mtcp);
1099 		}
1100 	}
1101 
1102 #if TESTING
1103 	DestroyRemainingFlows(mtcp);
1104 #endif
1105 
1106 	TRACE_DBG("MTCP thread %d out of main loop.\n", ctx->cpu);
1107 	/* flush logs */
1108 	flush_log_data(mtcp);
1109 	TRACE_DBG("MTCP thread %d flushed logs.\n", ctx->cpu);
1110 	InterruptApplication(mtcp);
1111 	TRACE_INFO("MTCP thread %d finished.\n", ctx->cpu);
1112 }
1113 /*----------------------------------------------------------------------------*/
1114 struct mtcp_sender *
1115 CreateMTCPSender(int ifidx)
1116 {
1117 	struct mtcp_sender *sender;
1118 
1119 	sender = (struct mtcp_sender *)calloc(1, sizeof(struct mtcp_sender));
1120 	if (!sender) {
1121 		return NULL;
1122 	}
1123 
1124 	sender->ifidx = ifidx;
1125 
1126 	TAILQ_INIT(&sender->control_list);
1127 	TAILQ_INIT(&sender->send_list);
1128 	TAILQ_INIT(&sender->ack_list);
1129 
1130 	sender->control_list_cnt = 0;
1131 	sender->send_list_cnt = 0;
1132 	sender->ack_list_cnt = 0;
1133 
1134 	return sender;
1135 }
1136 /*----------------------------------------------------------------------------*/
1137 void
1138 DestroyMTCPSender(struct mtcp_sender *sender)
1139 {
1140 	free(sender);
1141 }
1142 /*----------------------------------------------------------------------------*/
1143 static mtcp_manager_t
1144 InitializeMTCPManager(struct mtcp_thread_context* ctx)
1145 {
1146 	mtcp_manager_t mtcp;
1147 	char log_name[MAX_FILE_NAME];
1148 	int i;
1149 
1150 	posix_seq_srand((unsigned)pthread_self());
1151 
1152 	mtcp = (mtcp_manager_t)calloc(1, sizeof(struct mtcp_manager));
1153 	if (!mtcp) {
1154 		perror("malloc");
1155 		CTRACE_ERROR("Failed to allocate mtcp_manager.\n");
1156 		return NULL;
1157 	}
1158 	g_mtcp[ctx->cpu] = mtcp;
1159 
1160 	mtcp->tcp_flow_table = CreateHashtable();
1161 	if (!mtcp->tcp_flow_table) {
1162 		CTRACE_ERROR("Falied to allocate tcp flow table.\n");
1163 		return NULL;
1164 	}
1165 
1166 #ifdef HUGEPAGE
1167 #define	IS_HUGEPAGE 1
1168 #else
1169 #define	IS_HUGEPAGE 0
1170 #endif
1171 	if (mon_app_exists) {
1172 		/* initialize event callback */
1173 #ifdef NEWEV
1174 		InitEvent(mtcp);
1175 #else
1176 		InitEvent(mtcp, NUM_EV_TABLE);
1177 #endif
1178 	}
1179 
1180 	if (!(mtcp->bufseg_pool = MPCreate(sizeof(tcpbufseg_t),
1181 			sizeof(tcpbufseg_t) * g_config.mos->max_concurrency *
1182 			((g_config.mos->rmem_size - 1) / UNITBUFSIZE + 1), 0))) {
1183 		TRACE_ERROR("Failed to allocate ev_table pool\n");
1184 		exit(0);
1185 	}
1186 	if (!(mtcp->sockent_pool = MPCreate(sizeof(struct sockent),
1187 			sizeof(struct sockent) * g_config.mos->max_concurrency * 3, 0))) {
1188 		TRACE_ERROR("Failed to allocate ev_table pool\n");
1189 		exit(0);
1190 	}
1191 #ifdef USE_TIMER_POOL
1192 	if (!(mtcp->timer_pool = MPCreate(sizeof(struct timer),
1193 					  sizeof(struct timer) * g_config.mos->max_concurrency * 10, 0))) {
1194 		TRACE_ERROR("Failed to allocate ev_table pool\n");
1195 		exit(0);
1196 	}
1197 #endif
1198 	mtcp->flow_pool = MPCreate(sizeof(tcp_stream),
1199 								sizeof(tcp_stream) * g_config.mos->max_concurrency, IS_HUGEPAGE);
1200 	if (!mtcp->flow_pool) {
1201 		CTRACE_ERROR("Failed to allocate tcp flow pool.\n");
1202 		return NULL;
1203 	}
1204 	mtcp->rv_pool = MPCreate(sizeof(struct tcp_recv_vars),
1205 			sizeof(struct tcp_recv_vars) * g_config.mos->max_concurrency, IS_HUGEPAGE);
1206 	if (!mtcp->rv_pool) {
1207 		CTRACE_ERROR("Failed to allocate tcp recv variable pool.\n");
1208 		return NULL;
1209 	}
1210 	mtcp->sv_pool = MPCreate(sizeof(struct tcp_send_vars),
1211 			sizeof(struct tcp_send_vars) * g_config.mos->max_concurrency, IS_HUGEPAGE);
1212 	if (!mtcp->sv_pool) {
1213 		CTRACE_ERROR("Failed to allocate tcp send variable pool.\n");
1214 		return NULL;
1215 	}
1216 
1217 	mtcp->rbm_snd = SBManagerCreate(g_config.mos->wmem_size, g_config.mos->no_ring_buffers,
1218 					g_config.mos->max_concurrency);
1219 	if (!mtcp->rbm_snd) {
1220 		CTRACE_ERROR("Failed to create send ring buffer.\n");
1221 		return NULL;
1222 	}
1223 
1224 	mtcp->smap = (socket_map_t)calloc(g_config.mos->max_concurrency, sizeof(struct socket_map));
1225 	if (!mtcp->smap) {
1226 		perror("calloc");
1227 		CTRACE_ERROR("Failed to allocate memory for stream map.\n");
1228 		return NULL;
1229 	}
1230 
1231 	if (mon_app_exists) {
1232 		mtcp->msmap = (socket_map_t)calloc(g_config.mos->max_concurrency, sizeof(struct socket_map));
1233 		if (!mtcp->msmap) {
1234 			perror("calloc");
1235 			CTRACE_ERROR("Failed to allocate memory for monitor stream map.\n");
1236 			return NULL;
1237 		}
1238 
1239 		for (i = 0; i < g_config.mos->max_concurrency; i++) {
1240 			mtcp->msmap[i].monitor_stream = calloc(1, sizeof(struct mon_stream));
1241 			if (!mtcp->msmap[i].monitor_stream) {
1242 				perror("calloc");
1243 				CTRACE_ERROR("Failed to allocate memory for monitr stream map\n");
1244 				return NULL;
1245 			}
1246 		}
1247 	}
1248 
1249 	TAILQ_INIT(&mtcp->timer_list);
1250 	TAILQ_INIT(&mtcp->monitors);
1251 
1252 	TAILQ_INIT(&mtcp->free_smap);
1253 	for (i = 0; i < g_config.mos->max_concurrency; i++) {
1254 		mtcp->smap[i].id = i;
1255 		mtcp->smap[i].socktype = MOS_SOCK_UNUSED;
1256 		memset(&mtcp->smap[i].saddr, 0, sizeof(struct sockaddr_in));
1257 		mtcp->smap[i].stream = NULL;
1258 		TAILQ_INSERT_TAIL(&mtcp->free_smap, &mtcp->smap[i], link);
1259 	}
1260 
1261 	if (mon_app_exists) {
1262 		TAILQ_INIT(&mtcp->free_msmap);
1263 		for (i = 0; i < g_config.mos->max_concurrency; i++) {
1264 			mtcp->msmap[i].id = i;
1265 			mtcp->msmap[i].socktype = MOS_SOCK_UNUSED;
1266 			memset(&mtcp->msmap[i].saddr, 0, sizeof(struct sockaddr_in));
1267 			TAILQ_INSERT_TAIL(&mtcp->free_msmap, &mtcp->msmap[i], link);
1268 		}
1269 	}
1270 
1271 	mtcp->ctx = ctx;
1272 	mtcp->ep = NULL;
1273 
1274 	snprintf(log_name, MAX_FILE_NAME, "%s/"LOG_FILE_NAME"_%d",
1275 			g_config.mos->mos_log, ctx->cpu);
1276 	mtcp->log_fp = fopen(log_name, "w+");
1277 	if (!mtcp->log_fp) {
1278 		perror("fopen");
1279 		CTRACE_ERROR("Failed to create file for logging. (%s)\n", log_name);
1280 		return NULL;
1281 	}
1282 	mtcp->sp_fd = g_logctx[ctx->cpu]->pair_sp_fd;
1283 	mtcp->logger = g_logctx[ctx->cpu];
1284 
1285 	mtcp->connectq = CreateStreamQueue(BACKLOG_SIZE);
1286 	if (!mtcp->connectq) {
1287 		CTRACE_ERROR("Failed to create connect queue.\n");
1288 		return NULL;
1289 	}
1290 	mtcp->sendq = CreateStreamQueue(g_config.mos->max_concurrency);
1291 	if (!mtcp->sendq) {
1292 		CTRACE_ERROR("Failed to create send queue.\n");
1293 		return NULL;
1294 	}
1295 	mtcp->ackq = CreateStreamQueue(g_config.mos->max_concurrency);
1296 	if (!mtcp->ackq) {
1297 		CTRACE_ERROR("Failed to create ack queue.\n");
1298 		return NULL;
1299 	}
1300 	mtcp->closeq = CreateStreamQueue(g_config.mos->max_concurrency);
1301 	if (!mtcp->closeq) {
1302 		CTRACE_ERROR("Failed to create close queue.\n");
1303 		return NULL;
1304 	}
1305 	mtcp->closeq_int = CreateInternalStreamQueue(g_config.mos->max_concurrency);
1306 	if (!mtcp->closeq_int) {
1307 		CTRACE_ERROR("Failed to create close queue.\n");
1308 		return NULL;
1309 	}
1310 	mtcp->resetq = CreateStreamQueue(g_config.mos->max_concurrency);
1311 	if (!mtcp->resetq) {
1312 		CTRACE_ERROR("Failed to create reset queue.\n");
1313 		return NULL;
1314 	}
1315 	mtcp->resetq_int = CreateInternalStreamQueue(g_config.mos->max_concurrency);
1316 	if (!mtcp->resetq_int) {
1317 		CTRACE_ERROR("Failed to create reset queue.\n");
1318 		return NULL;
1319 	}
1320 	mtcp->destroyq = CreateStreamQueue(g_config.mos->max_concurrency);
1321 	if (!mtcp->destroyq) {
1322 		CTRACE_ERROR("Failed to create destroy queue.\n");
1323 		return NULL;
1324 	}
1325 
1326 	mtcp->g_sender = CreateMTCPSender(-1);
1327 	if (!mtcp->g_sender) {
1328 		CTRACE_ERROR("Failed to create global sender structure.\n");
1329 		return NULL;
1330 	}
1331 	for (i = 0; i < g_config.mos->netdev_table->num; i++) {
1332 		mtcp->n_sender[i] = CreateMTCPSender(i);
1333 		if (!mtcp->n_sender[i]) {
1334 			CTRACE_ERROR("Failed to create per-nic sender structure.\n");
1335 			return NULL;
1336 		}
1337 	}
1338 
1339 	mtcp->rto_store = InitRTOHashstore();
1340 	TAILQ_INIT(&mtcp->timewait_list);
1341 	TAILQ_INIT(&mtcp->timeout_list);
1342 
1343 	return mtcp;
1344 }
1345 /*----------------------------------------------------------------------------*/
1346 static void *
1347 MTCPRunThread(void *arg)
1348 {
1349 	mctx_t mctx = (mctx_t)arg;
1350 	int cpu = mctx->cpu;
1351 	int working;
1352 	struct mtcp_manager *mtcp;
1353 	struct mtcp_thread_context *ctx;
1354 
1355 	/* affinitize the thread to this core first */
1356 	mtcp_core_affinitize(cpu);
1357 
1358 	/* memory alloc after core affinitization would use local memory
1359 	   most time */
1360 	ctx = calloc(1, sizeof(*ctx));
1361 	if (!ctx) {
1362 		perror("calloc");
1363 		TRACE_ERROR("Failed to calloc mtcp context.\n");
1364 		exit(-1);
1365 	}
1366 	ctx->thread = pthread_self();
1367 	ctx->cpu = cpu;
1368 	mtcp = ctx->mtcp_manager = InitializeMTCPManager(ctx);
1369 	if (!mtcp) {
1370 		TRACE_ERROR("Failed to initialize mtcp manager.\n");
1371 		exit(-1);
1372 	}
1373 
1374 	/* assign mtcp context's underlying I/O module */
1375 	mtcp->iom = current_iomodule_func;
1376 
1377 	/* I/O initializing */
1378 	if (mtcp->iom->init_handle)
1379 		mtcp->iom->init_handle(ctx);
1380 
1381 	if (pthread_mutex_init(&ctx->flow_pool_lock, NULL)) {
1382 		perror("pthread_mutex_init of ctx->flow_pool_lock\n");
1383 		exit(-1);
1384 	}
1385 
1386 	if (pthread_mutex_init(&ctx->socket_pool_lock, NULL)) {
1387 		perror("pthread_mutex_init of ctx->socket_pool_lock\n");
1388 		exit(-1);
1389 	}
1390 
1391 	SQ_LOCK_INIT(&ctx->connect_lock, "ctx->connect_lock", exit(-1));
1392 	SQ_LOCK_INIT(&ctx->close_lock, "ctx->close_lock", exit(-1));
1393 	SQ_LOCK_INIT(&ctx->reset_lock, "ctx->reset_lock", exit(-1));
1394 	SQ_LOCK_INIT(&ctx->sendq_lock, "ctx->sendq_lock", exit(-1));
1395 	SQ_LOCK_INIT(&ctx->ackq_lock, "ctx->ackq_lock", exit(-1));
1396 	SQ_LOCK_INIT(&ctx->destroyq_lock, "ctx->destroyq_lock", exit(-1));
1397 
1398 	/* remember this context pointer for signal processing */
1399 	g_pctx[cpu] = ctx;
1400 	mlockall(MCL_CURRENT);
1401 
1402 	// attach (nic device, queue)
1403 	working = AttachDevice(ctx);
1404 	if (working != 0) {
1405 		sem_post(&g_init_sem[ctx->cpu]);
1406 		TRACE_DBG("MTCP thread %d finished. Not attached any device\n", ctx->cpu);
1407 		pthread_exit(NULL);
1408 	}
1409 
1410 	TRACE_DBG("CPU %d: initialization finished.\n", cpu);
1411 	sem_post(&g_init_sem[ctx->cpu]);
1412 
1413 	/* start the main loop */
1414 	RunMainLoop(ctx);
1415 
1416 	TRACE_DBG("MTCP thread %d finished.\n", ctx->cpu);
1417 
1418 	/* signaling mTCP thread is done */
1419 	sem_post(&g_done_sem[mctx->cpu]);
1420 
1421 	//pthread_exit(NULL);
1422 	return 0;
1423 }
1424 /*----------------------------------------------------------------------------*/
1425 #ifdef ENABLE_DPDK
1426 static int MTCPDPDKRunThread(void *arg)
1427 {
1428 	MTCPRunThread(arg);
1429 	return 0;
1430 }
1431 #endif /* !ENABLE_DPDK */
1432 /*----------------------------------------------------------------------------*/
1433 mctx_t
1434 mtcp_create_context(int cpu)
1435 {
1436 	mctx_t mctx;
1437 	int ret;
1438 
1439 	if (cpu >=  g_config.mos->num_cores) {
1440 		TRACE_ERROR("Failed initialize new mtcp context. "
1441 					"Requested cpu id %d exceed the number of cores %d configured to use.\n",
1442 					cpu, g_config.mos->num_cores);
1443 		return NULL;
1444 	}
1445 
1446         /* check if mtcp_create_context() was already initialized */
1447         if (g_logctx[cpu] != NULL) {
1448                 TRACE_ERROR("%s was already initialized before!\n",
1449                             __FUNCTION__);
1450                 return NULL;
1451         }
1452 
1453 	ret = sem_init(&g_init_sem[cpu], 0, 0);
1454 	if (ret) {
1455 		TRACE_ERROR("Failed initialize init_sem.\n");
1456 		return NULL;
1457 	}
1458 
1459 	ret = sem_init(&g_done_sem[cpu], 0, 0);
1460 	if (ret) {
1461 		TRACE_ERROR("Failed initialize done_sem.\n");
1462 		return NULL;
1463 	}
1464 
1465 	mctx = (mctx_t)calloc(1, sizeof(struct mtcp_context));
1466 	if (!mctx) {
1467 		TRACE_ERROR("Failed to allocate memory for mtcp_context.\n");
1468 		return NULL;
1469 	}
1470 	mctx->cpu = cpu;
1471 	g_ctx[cpu] = mctx;
1472 
1473 	/* initialize logger */
1474 	g_logctx[cpu] = (struct log_thread_context *)
1475 			calloc(1, sizeof(struct log_thread_context));
1476 	if (!g_logctx[cpu]) {
1477 		perror("malloc");
1478 		TRACE_ERROR("Failed to allocate memory for log thread context.\n");
1479 		return NULL;
1480 	}
1481 	InitLogThreadContext(g_logctx[cpu], cpu);
1482 	if (pthread_create(&log_thread[cpu],
1483 			   NULL, ThreadLogMain, (void *)g_logctx[cpu])) {
1484 		perror("pthread_create");
1485 		TRACE_ERROR("Failed to create log thread\n");
1486 		return NULL;
1487 	}
1488 
1489 #ifdef ENABLE_DPDK
1490 	/* Wake up mTCP threads (wake up I/O threads) */
1491 	if (current_iomodule_func == &dpdk_module_func) {
1492 		int master;
1493 		master = rte_get_master_lcore();
1494 		if (master == cpu) {
1495 			lcore_config[master].ret = 0;
1496 			lcore_config[master].state = FINISHED;
1497 			if (pthread_create(&g_thread[cpu],
1498 					   NULL, MTCPRunThread, (void *)mctx) != 0) {
1499 				TRACE_ERROR("pthread_create of mtcp thread failed!\n");
1500 				return NULL;
1501 			}
1502 		} else
1503 			rte_eal_remote_launch(MTCPDPDKRunThread, mctx, cpu);
1504 	} else
1505 #endif /* !ENABLE_DPDK */
1506 		{
1507 			if (pthread_create(&g_thread[cpu],
1508 					   NULL, MTCPRunThread, (void *)mctx) != 0) {
1509 				TRACE_ERROR("pthread_create of mtcp thread failed!\n");
1510 				return NULL;
1511 			}
1512 		}
1513 
1514 	sem_wait(&g_init_sem[cpu]);
1515 	sem_destroy(&g_init_sem[cpu]);
1516 
1517 	running[cpu] = TRUE;
1518 
1519 #ifdef NETSTAT
1520 #if NETSTAT_TOTAL
1521 	if (printer < 0) {
1522 		printer = cpu;
1523 		TRACE_INFO("CPU %d is in charge of printing stats.\n", printer);
1524 	}
1525 #endif
1526 #endif
1527 
1528 	return mctx;
1529 }
1530 /*----------------------------------------------------------------------------*/
1531 /**
1532  * TODO: It currently always returns 0. Add appropriate error return values
1533  */
1534 int
1535 mtcp_destroy_context(mctx_t mctx)
1536 {
1537 	struct mtcp_thread_context *ctx = g_pctx[mctx->cpu];
1538 	struct mtcp_manager *mtcp = ctx->mtcp_manager;
1539 	struct log_thread_context *log_ctx = mtcp->logger;
1540 	int ret, i;
1541 
1542 	TRACE_DBG("CPU %d: mtcp_destroy_context()\n", mctx->cpu);
1543 
1544 	/* close all stream sockets that are still open */
1545 	if (!ctx->exit) {
1546 		for (i = 0; i < g_config.mos->max_concurrency; i++) {
1547 			if (mtcp->smap[i].socktype == MOS_SOCK_STREAM) {
1548 				TRACE_DBG("Closing remaining socket %d (%s)\n",
1549 						i, TCPStateToString(mtcp->smap[i].stream));
1550 				DumpStream(mtcp, mtcp->smap[i].stream);
1551 				mtcp_close(mctx, i);
1552 			}
1553 		}
1554 	}
1555 
1556 	ctx->done = 1;
1557 
1558 	//pthread_kill(g_thread[mctx->cpu], SIGINT);
1559 #ifdef ENABLE_DPDK
1560 	ctx->exit = 1;
1561 	/* XXX - dpdk logic changes */
1562 	if (current_iomodule_func == &dpdk_module_func) {
1563 		int master = rte_get_master_lcore();
1564 		if (master == mctx->cpu)
1565 			pthread_join(g_thread[mctx->cpu], NULL);
1566 		else
1567 			rte_eal_wait_lcore(mctx->cpu);
1568 	} else
1569 #endif /* !ENABLE_DPDK */
1570 		{
1571 			pthread_join(g_thread[mctx->cpu], NULL);
1572 		}
1573 
1574 	TRACE_INFO("MTCP thread %d joined.\n", mctx->cpu);
1575 	running[mctx->cpu] = FALSE;
1576 
1577 #ifdef NETSTAT
1578 #if NETSTAT_TOTAL
1579 	if (printer == mctx->cpu) {
1580 		for (i = 0; i < num_cpus; i++) {
1581 			if (i != mctx->cpu && running[i]) {
1582 				printer = i;
1583 				break;
1584 			}
1585 		}
1586 	}
1587 #endif
1588 #endif
1589 
1590 	log_ctx->done = 1;
1591 	ret = write(log_ctx->pair_sp_fd, "F", 1);
1592 	if (ret != 1)
1593 		TRACE_ERROR("CPU %d: Fail to signal socket pair\n", mctx->cpu);
1594 
1595 	pthread_join(log_thread[ctx->cpu], NULL);
1596 	fclose(mtcp->log_fp);
1597 	TRACE_LOG("Log thread %d joined.\n", mctx->cpu);
1598 
1599 	if (mtcp->connectq) {
1600 		DestroyStreamQueue(mtcp->connectq);
1601 		mtcp->connectq = NULL;
1602 	}
1603 	if (mtcp->sendq) {
1604 		DestroyStreamQueue(mtcp->sendq);
1605 		mtcp->sendq = NULL;
1606 	}
1607 	if (mtcp->ackq) {
1608 		DestroyStreamQueue(mtcp->ackq);
1609 		mtcp->ackq = NULL;
1610 	}
1611 	if (mtcp->closeq) {
1612 		DestroyStreamQueue(mtcp->closeq);
1613 		mtcp->closeq = NULL;
1614 	}
1615 	if (mtcp->closeq_int) {
1616 		DestroyInternalStreamQueue(mtcp->closeq_int);
1617 		mtcp->closeq_int = NULL;
1618 	}
1619 	if (mtcp->resetq) {
1620 		DestroyStreamQueue(mtcp->resetq);
1621 		mtcp->resetq = NULL;
1622 	}
1623 	if (mtcp->resetq_int) {
1624 		DestroyInternalStreamQueue(mtcp->resetq_int);
1625 		mtcp->resetq_int = NULL;
1626 	}
1627 	if (mtcp->destroyq) {
1628 		DestroyStreamQueue(mtcp->destroyq);
1629 		mtcp->destroyq = NULL;
1630 	}
1631 
1632 	DestroyMTCPSender(mtcp->g_sender);
1633 	for (i = 0; i < g_config.mos->netdev_table->num; i++) {
1634 		DestroyMTCPSender(mtcp->n_sender[i]);
1635 	}
1636 
1637 	MPDestroy(mtcp->rv_pool);
1638 	MPDestroy(mtcp->sv_pool);
1639 	MPDestroy(mtcp->flow_pool);
1640 
1641 	if (mtcp->ap) {
1642 		DestroyAddressPool(mtcp->ap);
1643 	}
1644 
1645 	SQ_LOCK_DESTROY(&ctx->connect_lock);
1646 	SQ_LOCK_DESTROY(&ctx->close_lock);
1647 	SQ_LOCK_DESTROY(&ctx->reset_lock);
1648 	SQ_LOCK_DESTROY(&ctx->sendq_lock);
1649 	SQ_LOCK_DESTROY(&ctx->ackq_lock);
1650 	SQ_LOCK_DESTROY(&ctx->destroyq_lock);
1651 
1652 	//TRACE_INFO("MTCP thread %d destroyed.\n", mctx->cpu);
1653 	if (mtcp->iom->destroy_handle)
1654 		mtcp->iom->destroy_handle(ctx);
1655 	free(ctx);
1656 	free(mctx);
1657 
1658 	return 0;
1659 }
1660 /*----------------------------------------------------------------------------*/
1661 mtcp_sighandler_t
1662 mtcp_register_signal(int signum, mtcp_sighandler_t handler)
1663 {
1664 	mtcp_sighandler_t prev;
1665 
1666 	if (signum == SIGINT) {
1667 		prev = app_signal_handler;
1668 		app_signal_handler = handler;
1669 	} else {
1670 		if ((prev = signal(signum, handler)) == SIG_ERR) {
1671 			perror("signal");
1672 			return SIG_ERR;
1673 		}
1674 	}
1675 
1676 	return prev;
1677 }
1678 /*----------------------------------------------------------------------------*/
1679 int
1680 mtcp_getconf(struct mtcp_conf *conf)
1681 {
1682 	int i, j;
1683 
1684 	if (!conf)
1685 		return -1;
1686 
1687 	conf->num_cores = g_config.mos->num_cores;
1688 	conf->max_concurrency = g_config.mos->max_concurrency;
1689 	conf->cpu_mask = g_config.mos->cpu_mask;
1690 
1691 	conf->rcvbuf_size = g_config.mos->rmem_size;
1692 	conf->sndbuf_size = g_config.mos->wmem_size;
1693 
1694 	conf->tcp_timewait = g_config.mos->tcp_tw_interval;
1695 	conf->tcp_timeout = g_config.mos->tcp_timeout;
1696 
1697 	i = 0;
1698 	struct conf_block *bwalk;
1699 	TAILQ_FOREACH(bwalk, &g_config.app_blkh, link) {
1700 		struct app_conf *app_conf = (struct app_conf *)bwalk->conf;
1701 		for (j = 0; j < app_conf->app_argc; j++)
1702 			conf->app_argv[i][j] = app_conf->app_argv[j];
1703 		conf->app_argc[i] = app_conf->app_argc;
1704 		conf->app_cpu_mask[i] = app_conf->cpu_mask;
1705 		i++;
1706 	}
1707 	conf->num_app = i;
1708 
1709 	return 0;
1710 }
1711 /*----------------------------------------------------------------------------*/
1712 int
1713 mtcp_setconf(const struct mtcp_conf *conf)
1714 {
1715 	if (!conf)
1716 		return -1;
1717 
1718 	g_config.mos->num_cores = conf->num_cores;
1719 	g_config.mos->max_concurrency = conf->max_concurrency;
1720 
1721 	g_config.mos->rmem_size = conf->rcvbuf_size;
1722 	g_config.mos->wmem_size = conf->sndbuf_size;
1723 
1724 	g_config.mos->tcp_tw_interval = conf->tcp_timewait;
1725 	g_config.mos->tcp_timeout = conf->tcp_timeout;
1726 
1727 	TRACE_CONFIG("Configuration updated by mtcp_setconf().\n");
1728 	//PrintConfiguration();
1729 
1730 	return 0;
1731 }
1732 /*----------------------------------------------------------------------------*/
1733 int
1734 mtcp_init(const char *config_file)
1735 {
1736 	int i;
1737 	int ret;
1738 
1739 	if (geteuid()) {
1740 		TRACE_CONFIG("[CAUTION] Run as root if mlock is necessary.\n");
1741 	}
1742 
1743 	/* getting cpu and NIC */
1744 	num_cpus = GetNumCPUs();
1745 	assert(num_cpus >= 1);
1746 	for (i = 0; i < num_cpus; i++) {
1747 		g_mtcp[i] = NULL;
1748 		running[i] = FALSE;
1749 		sigint_cnt[i] = 0;
1750 	}
1751 
1752 	ret = LoadConfigurationUpperHalf(config_file);
1753 	if (ret) {
1754 		TRACE_CONFIG("Error occured while loading configuration.\n");
1755 		return -1;
1756 	}
1757 
1758 #if defined(ENABLE_PSIO)
1759 	current_iomodule_func = &ps_module_func;
1760 #elif defined(ENABLE_DPDK)
1761 	current_iomodule_func = &dpdk_module_func;
1762 #elif defined(ENABLE_PCAP)
1763 	current_iomodule_func = &pcap_module_func;
1764 #elif defined(ENABLE_DPDKR)
1765 	current_iomodule_func = &dpdkr_module_func;
1766 #elif defined(ENABLE_NETMAP)
1767 	current_iomodule_func = &netmap_module_func;
1768 #endif
1769 
1770 	if (current_iomodule_func->load_module_upper_half)
1771 		current_iomodule_func->load_module_upper_half();
1772 
1773 	LoadConfigurationLowerHalf();
1774 
1775 	//PrintConfiguration();
1776 
1777 	/* TODO: this should be fixed */
1778 	ap = CreateAddressPool(g_config.mos->netdev_table->ent[0]->ip_addr, 1);
1779 	if (!ap) {
1780 		TRACE_CONFIG("Error occured while creating address pool.\n");
1781 		return -1;
1782 	}
1783 
1784 	//PrintInterfaceInfo();
1785 	//PrintRoutingTable();
1786 	//PrintARPTable();
1787 	InitARPTable();
1788 
1789 	if (signal(SIGUSR1, HandleSignal) == SIG_ERR) {
1790 		perror("signal, SIGUSR1");
1791 		return -1;
1792 	}
1793 	if (signal(SIGINT, HandleSignal) == SIG_ERR) {
1794 		perror("signal, SIGINT");
1795 		return -1;
1796 	}
1797 	app_signal_handler = NULL;
1798 
1799 	printf("load_module(): %p\n", current_iomodule_func);
1800 	/* load system-wide io module specs */
1801 	if (current_iomodule_func->load_module_lower_half)
1802 		current_iomodule_func->load_module_lower_half();
1803 
1804 	GlobInitEvent();
1805 
1806 	PrintConf(&g_config);
1807 
1808 	return 0;
1809 }
1810 /*----------------------------------------------------------------------------*/
1811 int
1812 mtcp_destroy()
1813 {
1814 	int i;
1815 
1816 	/* wait until all threads are closed */
1817 	for (i = 0; i < num_cpus; i++) {
1818 		if (running[i]) {
1819 			if (pthread_join(g_thread[i], NULL) != 0)
1820 				return -1;
1821 		}
1822 	}
1823 
1824 	DestroyAddressPool(ap);
1825 
1826 	TRACE_INFO("All MTCP threads are joined.\n");
1827 
1828 	return 0;
1829 }
1830 /*----------------------------------------------------------------------------*/
1831