xref: /mOS-networking-stack/core/src/core.c (revision df3fae06)
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 #ifdef DUMP_STREAM
790 			DumpStream(mtcp, walk);
791 #endif
792 			DestroyTCPStream(mtcp, walk);
793 			cnt++;
794 		}
795 	}
796 
797 	return cnt;
798 }
799 #endif
800 /*----------------------------------------------------------------------------*/
801 static void
802 InterruptApplication(mtcp_manager_t mtcp)
803 {
804 	/* interrupt if the mtcp_epoll_wait() is waiting */
805 	if (mtcp->ep) {
806 		pthread_mutex_lock(&mtcp->ep->epoll_lock);
807 		if (mtcp->ep->waiting) {
808 			pthread_cond_signal(&mtcp->ep->epoll_cond);
809 		}
810 		pthread_mutex_unlock(&mtcp->ep->epoll_lock);
811 	}
812 	/* interrupt if the accept() is waiting */
813 	if (mtcp->listener) {
814 		if (mtcp->listener->socket) {
815 			pthread_mutex_lock(&mtcp->listener->accept_lock);
816 			if (!(mtcp->listener->socket->opts & MTCP_NONBLOCK)) {
817 				pthread_cond_signal(&mtcp->listener->accept_cond);
818 			}
819 			pthread_mutex_unlock(&mtcp->listener->accept_lock);
820 		}
821 	}
822 }
823 /*----------------------------------------------------------------------------*/
824 void
825 RunPassiveLoop(mtcp_manager_t mtcp)
826 {
827 	sem_wait(&g_done_sem[mtcp->ctx->cpu]);
828 	sem_destroy(&g_done_sem[mtcp->ctx->cpu]);
829 	return;
830 }
831 /*----------------------------------------------------------------------------*/
832 static void
833 RunMainLoop(struct mtcp_thread_context *ctx)
834 {
835 	mtcp_manager_t mtcp = ctx->mtcp_manager;
836 	int i;
837 	int recv_cnt;
838 
839 #if E_PSIO
840 	int rx_inf, tx_inf;
841 #endif
842 
843 	struct timeval cur_ts = {0};
844 	uint32_t ts, ts_prev;
845 
846 #if TIME_STAT
847 	struct timeval prev_ts, processing_ts, tcheck_ts,
848 				   epoll_ts, handle_ts, xmit_ts, select_ts;
849 #endif
850 	int thresh;
851 
852 	gettimeofday(&cur_ts, NULL);
853 
854 #if E_PSIO
855 	/* do nothing */
856 #else
857 #if !USE_CHUNK_BUF
858 	/* create packet write chunk */
859 	InitWriteChunks(handle, ctx->w_chunk);
860 	for (i = 0; i < ETH_NUM; i++) {
861 		ctx->w_chunk[i].cnt = 0;
862 		ctx->w_off[i] = 0;
863 		ctx->w_cur_idx[i] = 0;
864 	}
865 #endif
866 #endif
867 
868 	TRACE_DBG("CPU %d: mtcp thread running.\n", ctx->cpu);
869 
870 #if TIME_STAT
871 	prev_ts = cur_ts;
872 	InitStatCounter(&mtcp->rtstat.round);
873 	InitStatCounter(&mtcp->rtstat.processing);
874 	InitStatCounter(&mtcp->rtstat.tcheck);
875 	InitStatCounter(&mtcp->rtstat.epoll);
876 	InitStatCounter(&mtcp->rtstat.handle);
877 	InitStatCounter(&mtcp->rtstat.xmit);
878 	InitStatCounter(&mtcp->rtstat.select);
879 #endif
880 
881 	ts = ts_prev = 0;
882 	while ((!ctx->done || mtcp->flow_cnt) && !ctx->exit) {
883 
884 		STAT_COUNT(mtcp->runstat.rounds);
885 		recv_cnt = 0;
886 
887 #if E_PSIO
888 #if 0
889 		event.timeout = PS_SELECT_TIMEOUT;
890 		NID_ZERO(event.rx_nids);
891 		NID_ZERO(event.tx_nids);
892 		NID_ZERO(rx_avail);
893 		//NID_ZERO(tx_avail);
894 #endif
895 		gettimeofday(&cur_ts, NULL);
896 #if TIME_STAT
897 		/* measure the inter-round delay */
898 		UpdateStatCounter(&mtcp->rtstat.round, TimeDiffUs(&cur_ts, &prev_ts));
899 		prev_ts = cur_ts;
900 #endif
901 
902 		ts = TIMEVAL_TO_TS(&cur_ts);
903 		mtcp->cur_ts = ts;
904 
905 		for (rx_inf = 0; rx_inf < g_config.mos->netdev_table->num; rx_inf++) {
906 
907 			recv_cnt = mtcp->iom->recv_pkts(ctx, rx_inf);
908 			STAT_COUNT(mtcp->runstat.rounds_rx_try);
909 
910 			for (i = 0; i < recv_cnt; i++) {
911 				uint16_t len;
912 				uint8_t *pktbuf;
913 				pktbuf = mtcp->iom->get_rptr(mtcp->ctx, rx_inf, i, &len);
914 				ProcessPacket(mtcp, rx_inf, i, ts, pktbuf, len);
915 			}
916 
917 #ifdef ENABLE_DPDKR
918 			mtcp->iom->send_pkts(ctx, rx_inf);
919 			continue;
920 #endif
921 		}
922 		STAT_COUNT(mtcp->runstat.rounds_rx);
923 
924 #else /* E_PSIO */
925 		gettimeofday(&cur_ts, NULL);
926 		ts = TIMEVAL_TO_TS(&cur_ts);
927 		mtcp->cur_ts = ts;
928 		/*
929 		 * Read packets into a chunk from NIC
930 		 * chk_w_idx : next chunk index to write packets from NIC
931 		 */
932 
933 		STAT_COUNT(mtcp->runstat.rounds_rx_try);
934 		chunk.cnt = PS_CHUNK_SIZE;
935 		recv_cnt = ps_recv_chunk(handle, &chunk);
936 		if (recv_cnt < 0) {
937 			if (errno != EAGAIN && errno != EINTR) {
938 				perror("ps_recv_chunk");
939 				assert(0);
940 			}
941 		}
942 
943 		/*
944 		 * Handling Packets
945 		 * chk_r_idx : next chunk index to read and handle
946 		*/
947 		for (i = 0; i < recv_cnt; i++) {
948 			ProcessPacket(mtcp, chunk.queue.ifindex, ts,
949 					(u_char *)(chunk.buf + chunk.info[i].offset),
950 					chunk.info[i].len);
951 		}
952 
953 		if (recv_cnt > 0)
954 			STAT_COUNT(mtcp->runstat.rounds_rx);
955 #endif /* E_PSIO */
956 #if TIME_STAT
957 		gettimeofday(&processing_ts, NULL);
958 		UpdateStatCounter(&mtcp->rtstat.processing,
959 				TimeDiffUs(&processing_ts, &cur_ts));
960 #endif /* TIME_STAT */
961 
962 		/* Handle user defined timeout */
963 		struct timer *walk, *tmp;
964 		for (walk = TAILQ_FIRST(&mtcp->timer_list); walk != NULL; walk = tmp) {
965 			tmp = TAILQ_NEXT(walk, timer_link);
966 			if (TIMEVAL_LT(&cur_ts, &walk->exp))
967 				break;
968 
969 			struct mtcp_context mctx = {.cpu = ctx->cpu};
970 			walk->cb(&mctx, walk->id, 0, 0 /* FIXME */, NULL);
971 			DelTimer(mtcp, walk);
972 		}
973 
974 		/* interaction with application */
975 		if (mtcp->flow_cnt > 0) {
976 
977 			/* check retransmission timeout and timewait expire */
978 #if 0
979 			thresh = (int)mtcp->flow_cnt / (TS_TO_USEC(PER_STREAM_TCHECK));
980 			assert(thresh >= 0);
981 			if (thresh == 0)
982 				thresh = 1;
983 			if (recv_cnt > 0 && thresh > recv_cnt)
984 				thresh = recv_cnt;
985 #else
986 			thresh = g_config.mos->max_concurrency;
987 #endif
988 
989 			/* Eunyoung, you may fix this later
990 			 * if there is no rcv packet, we will send as much as possible
991 			 */
992 			if (thresh == -1)
993 				thresh = g_config.mos->max_concurrency;
994 
995 			CheckRtmTimeout(mtcp, ts, thresh);
996 			CheckTimewaitExpire(mtcp, ts, thresh);
997 
998 			if (g_config.mos->tcp_timeout > 0 && ts != ts_prev) {
999 				CheckConnectionTimeout(mtcp, ts, thresh);
1000 			}
1001 
1002 #if TIME_STAT
1003 		}
1004 		gettimeofday(&tcheck_ts, NULL);
1005 		UpdateStatCounter(&mtcp->rtstat.tcheck,
1006 				TimeDiffUs(&tcheck_ts, &processing_ts));
1007 
1008 		if (mtcp->flow_cnt > 0) {
1009 #endif /* TIME_STAT */
1010 
1011 		}
1012 
1013 		/*
1014 		 * before flushing epoll events, call monitor events for
1015 		 * all registered `read` events
1016 		 */
1017 		if (mtcp->num_msp > 0)
1018 			/* call this when only a standalone monitor is running */
1019 			FlushMonitorReadEvents(mtcp);
1020 
1021 		/* if epoll is in use, flush all the queued events */
1022 		if (mtcp->ep) {
1023 			FlushBufferedReadEvents(mtcp);
1024 			FlushEpollEvents(mtcp, ts);
1025 		}
1026 #if TIME_STAT
1027 		gettimeofday(&epoll_ts, NULL);
1028 		UpdateStatCounter(&mtcp->rtstat.epoll,
1029 				TimeDiffUs(&epoll_ts, &tcheck_ts));
1030 #endif /* TIME_STAT */
1031 
1032 		if (end_app_exists && mtcp->flow_cnt > 0) {
1033 			/* handle stream queues  */
1034 			HandleApplicationCalls(mtcp, ts);
1035 		}
1036 
1037 #ifdef ENABLE_DPDKR
1038 		continue;
1039 #endif
1040 
1041 #if TIME_STAT
1042 		gettimeofday(&handle_ts, NULL);
1043 		UpdateStatCounter(&mtcp->rtstat.handle,
1044 				TimeDiffUs(&handle_ts, &epoll_ts));
1045 #endif /* TIME_STAT */
1046 
1047 		WritePacketsToChunks(mtcp, ts);
1048 
1049 		/* send packets from write buffer */
1050 #if E_PSIO
1051 		/* With E_PSIO, send until tx is available */
1052 		int num_dev = g_config.mos->netdev_table->num;
1053 		if (likely(mtcp->iom->send_pkts != NULL))
1054 			for (tx_inf = 0; tx_inf < num_dev; tx_inf++) {
1055 				mtcp->iom->send_pkts(ctx, tx_inf);
1056 			}
1057 
1058 #else /* E_PSIO */
1059 		/* Without E_PSIO, try send chunks immediately */
1060 		for (i = 0; i < g_config.mos->netdev_table->num; i++) {
1061 #if USE_CHUNK_BUF
1062 			/* in the case of using ps_send_chunk_buf() without E_PSIO */
1063 			ret = FlushSendChunkBuf(mtcp, i);
1064 #else
1065 			/* if not using ps_send_chunk_buf() */
1066 			ret = FlushWriteBuffer(ctx, i);
1067 #endif
1068 			if (ret < 0) {
1069 				TRACE_ERROR("Failed to send chunks.\n");
1070 			} else if (ret > 0) {
1071 				STAT_COUNT(mtcp->runstat.rounds_tx);
1072 			}
1073 		}
1074 #endif /* E_PSIO */
1075 
1076 #if TIME_STAT
1077 		gettimeofday(&xmit_ts, NULL);
1078 		UpdateStatCounter(&mtcp->rtstat.xmit,
1079 				TimeDiffUs(&xmit_ts, &handle_ts));
1080 #endif /* TIME_STAT */
1081 
1082 		if (ts != ts_prev) {
1083 			ts_prev = ts;
1084 #ifdef NETSTAT
1085 			if (ctx->cpu == printer) {
1086 #ifdef RUN_ARP
1087 				ARPTimer(mtcp, ts);
1088 #endif
1089 #ifdef NETSTAT
1090 				PrintNetworkStats(mtcp, ts);
1091 #endif
1092 			}
1093 #endif /* NETSTAT */
1094 		}
1095 
1096 		if (mtcp->iom->select)
1097 			mtcp->iom->select(ctx);
1098 
1099 		if (ctx->interrupt) {
1100 			InterruptApplication(mtcp);
1101 		}
1102 	}
1103 
1104 #if TESTING
1105 	DestroyRemainingFlows(mtcp);
1106 #endif
1107 
1108 	TRACE_DBG("MTCP thread %d out of main loop.\n", ctx->cpu);
1109 	/* flush logs */
1110 	flush_log_data(mtcp);
1111 	TRACE_DBG("MTCP thread %d flushed logs.\n", ctx->cpu);
1112 	InterruptApplication(mtcp);
1113 	TRACE_INFO("MTCP thread %d finished.\n", ctx->cpu);
1114 }
1115 /*----------------------------------------------------------------------------*/
1116 struct mtcp_sender *
1117 CreateMTCPSender(int ifidx)
1118 {
1119 	struct mtcp_sender *sender;
1120 
1121 	sender = (struct mtcp_sender *)calloc(1, sizeof(struct mtcp_sender));
1122 	if (!sender) {
1123 		return NULL;
1124 	}
1125 
1126 	sender->ifidx = ifidx;
1127 
1128 	TAILQ_INIT(&sender->control_list);
1129 	TAILQ_INIT(&sender->send_list);
1130 	TAILQ_INIT(&sender->ack_list);
1131 
1132 	sender->control_list_cnt = 0;
1133 	sender->send_list_cnt = 0;
1134 	sender->ack_list_cnt = 0;
1135 
1136 	return sender;
1137 }
1138 /*----------------------------------------------------------------------------*/
1139 void
1140 DestroyMTCPSender(struct mtcp_sender *sender)
1141 {
1142 	free(sender);
1143 }
1144 /*----------------------------------------------------------------------------*/
1145 static mtcp_manager_t
1146 InitializeMTCPManager(struct mtcp_thread_context* ctx)
1147 {
1148 	mtcp_manager_t mtcp;
1149 	char log_name[MAX_FILE_NAME];
1150 	int i;
1151 
1152 	posix_seq_srand((unsigned)pthread_self());
1153 
1154 	mtcp = (mtcp_manager_t)calloc(1, sizeof(struct mtcp_manager));
1155 	if (!mtcp) {
1156 		perror("malloc");
1157 		CTRACE_ERROR("Failed to allocate mtcp_manager.\n");
1158 		return NULL;
1159 	}
1160 	g_mtcp[ctx->cpu] = mtcp;
1161 
1162 	mtcp->tcp_flow_table = CreateHashtable();
1163 	if (!mtcp->tcp_flow_table) {
1164 		CTRACE_ERROR("Falied to allocate tcp flow table.\n");
1165 		return NULL;
1166 	}
1167 
1168 #ifdef HUGEPAGE
1169 #define	IS_HUGEPAGE 1
1170 #else
1171 #define	IS_HUGEPAGE 0
1172 #endif
1173 	if (mon_app_exists) {
1174 		/* initialize event callback */
1175 #ifdef NEWEV
1176 		InitEvent(mtcp);
1177 #else
1178 		InitEvent(mtcp, NUM_EV_TABLE);
1179 #endif
1180 	}
1181 
1182 	if (!(mtcp->bufseg_pool = MPCreate(sizeof(tcpbufseg_t),
1183 			sizeof(tcpbufseg_t) * g_config.mos->max_concurrency *
1184 			((g_config.mos->rmem_size - 1) / UNITBUFSIZE + 1), 0))) {
1185 		TRACE_ERROR("Failed to allocate ev_table pool\n");
1186 		exit(0);
1187 	}
1188 	if (!(mtcp->sockent_pool = MPCreate(sizeof(struct sockent),
1189 			sizeof(struct sockent) * g_config.mos->max_concurrency * 3, 0))) {
1190 		TRACE_ERROR("Failed to allocate ev_table pool\n");
1191 		exit(0);
1192 	}
1193 #ifdef USE_TIMER_POOL
1194 	if (!(mtcp->timer_pool = MPCreate(sizeof(struct timer),
1195 					  sizeof(struct timer) * g_config.mos->max_concurrency * 10, 0))) {
1196 		TRACE_ERROR("Failed to allocate ev_table pool\n");
1197 		exit(0);
1198 	}
1199 #endif
1200 	mtcp->flow_pool = MPCreate(sizeof(tcp_stream),
1201 								sizeof(tcp_stream) * g_config.mos->max_concurrency, IS_HUGEPAGE);
1202 	if (!mtcp->flow_pool) {
1203 		CTRACE_ERROR("Failed to allocate tcp flow pool.\n");
1204 		return NULL;
1205 	}
1206 	mtcp->rv_pool = MPCreate(sizeof(struct tcp_recv_vars),
1207 			sizeof(struct tcp_recv_vars) * g_config.mos->max_concurrency, IS_HUGEPAGE);
1208 	if (!mtcp->rv_pool) {
1209 		CTRACE_ERROR("Failed to allocate tcp recv variable pool.\n");
1210 		return NULL;
1211 	}
1212 	mtcp->sv_pool = MPCreate(sizeof(struct tcp_send_vars),
1213 			sizeof(struct tcp_send_vars) * g_config.mos->max_concurrency, IS_HUGEPAGE);
1214 	if (!mtcp->sv_pool) {
1215 		CTRACE_ERROR("Failed to allocate tcp send variable pool.\n");
1216 		return NULL;
1217 	}
1218 
1219 	mtcp->rbm_snd = SBManagerCreate(g_config.mos->wmem_size, g_config.mos->no_ring_buffers,
1220 					g_config.mos->max_concurrency);
1221 	if (!mtcp->rbm_snd) {
1222 		CTRACE_ERROR("Failed to create send ring buffer.\n");
1223 		return NULL;
1224 	}
1225 
1226 	mtcp->smap = (socket_map_t)calloc(g_config.mos->max_concurrency, sizeof(struct socket_map));
1227 	if (!mtcp->smap) {
1228 		perror("calloc");
1229 		CTRACE_ERROR("Failed to allocate memory for stream map.\n");
1230 		return NULL;
1231 	}
1232 
1233 	if (mon_app_exists) {
1234 		mtcp->msmap = (socket_map_t)calloc(g_config.mos->max_concurrency, sizeof(struct socket_map));
1235 		if (!mtcp->msmap) {
1236 			perror("calloc");
1237 			CTRACE_ERROR("Failed to allocate memory for monitor stream map.\n");
1238 			return NULL;
1239 		}
1240 
1241 		for (i = 0; i < g_config.mos->max_concurrency; i++) {
1242 			mtcp->msmap[i].monitor_stream = calloc(1, sizeof(struct mon_stream));
1243 			if (!mtcp->msmap[i].monitor_stream) {
1244 				perror("calloc");
1245 				CTRACE_ERROR("Failed to allocate memory for monitr stream map\n");
1246 				return NULL;
1247 			}
1248 		}
1249 	}
1250 
1251 	TAILQ_INIT(&mtcp->timer_list);
1252 	TAILQ_INIT(&mtcp->monitors);
1253 
1254 	TAILQ_INIT(&mtcp->free_smap);
1255 	for (i = 0; i < g_config.mos->max_concurrency; i++) {
1256 		mtcp->smap[i].id = i;
1257 		mtcp->smap[i].socktype = MOS_SOCK_UNUSED;
1258 		memset(&mtcp->smap[i].saddr, 0, sizeof(struct sockaddr_in));
1259 		mtcp->smap[i].stream = NULL;
1260 		TAILQ_INSERT_TAIL(&mtcp->free_smap, &mtcp->smap[i], link);
1261 	}
1262 
1263 	if (mon_app_exists) {
1264 		TAILQ_INIT(&mtcp->free_msmap);
1265 		for (i = 0; i < g_config.mos->max_concurrency; i++) {
1266 			mtcp->msmap[i].id = i;
1267 			mtcp->msmap[i].socktype = MOS_SOCK_UNUSED;
1268 			memset(&mtcp->msmap[i].saddr, 0, sizeof(struct sockaddr_in));
1269 			TAILQ_INSERT_TAIL(&mtcp->free_msmap, &mtcp->msmap[i], link);
1270 		}
1271 	}
1272 
1273 	mtcp->ctx = ctx;
1274 	mtcp->ep = NULL;
1275 
1276 	snprintf(log_name, MAX_FILE_NAME, "%s/"LOG_FILE_NAME"_%d",
1277 			g_config.mos->mos_log, ctx->cpu);
1278 	mtcp->log_fp = fopen(log_name, "w+");
1279 	if (!mtcp->log_fp) {
1280 		perror("fopen");
1281 		CTRACE_ERROR("Failed to create file for logging. (%s)\n", log_name);
1282 		return NULL;
1283 	}
1284 	mtcp->sp_fd = g_logctx[ctx->cpu]->pair_sp_fd;
1285 	mtcp->logger = g_logctx[ctx->cpu];
1286 
1287 	mtcp->connectq = CreateStreamQueue(BACKLOG_SIZE);
1288 	if (!mtcp->connectq) {
1289 		CTRACE_ERROR("Failed to create connect queue.\n");
1290 		return NULL;
1291 	}
1292 	mtcp->sendq = CreateStreamQueue(g_config.mos->max_concurrency);
1293 	if (!mtcp->sendq) {
1294 		CTRACE_ERROR("Failed to create send queue.\n");
1295 		return NULL;
1296 	}
1297 	mtcp->ackq = CreateStreamQueue(g_config.mos->max_concurrency);
1298 	if (!mtcp->ackq) {
1299 		CTRACE_ERROR("Failed to create ack queue.\n");
1300 		return NULL;
1301 	}
1302 	mtcp->closeq = CreateStreamQueue(g_config.mos->max_concurrency);
1303 	if (!mtcp->closeq) {
1304 		CTRACE_ERROR("Failed to create close queue.\n");
1305 		return NULL;
1306 	}
1307 	mtcp->closeq_int = CreateInternalStreamQueue(g_config.mos->max_concurrency);
1308 	if (!mtcp->closeq_int) {
1309 		CTRACE_ERROR("Failed to create close queue.\n");
1310 		return NULL;
1311 	}
1312 	mtcp->resetq = CreateStreamQueue(g_config.mos->max_concurrency);
1313 	if (!mtcp->resetq) {
1314 		CTRACE_ERROR("Failed to create reset queue.\n");
1315 		return NULL;
1316 	}
1317 	mtcp->resetq_int = CreateInternalStreamQueue(g_config.mos->max_concurrency);
1318 	if (!mtcp->resetq_int) {
1319 		CTRACE_ERROR("Failed to create reset queue.\n");
1320 		return NULL;
1321 	}
1322 	mtcp->destroyq = CreateStreamQueue(g_config.mos->max_concurrency);
1323 	if (!mtcp->destroyq) {
1324 		CTRACE_ERROR("Failed to create destroy queue.\n");
1325 		return NULL;
1326 	}
1327 
1328 	mtcp->g_sender = CreateMTCPSender(-1);
1329 	if (!mtcp->g_sender) {
1330 		CTRACE_ERROR("Failed to create global sender structure.\n");
1331 		return NULL;
1332 	}
1333 	for (i = 0; i < g_config.mos->netdev_table->num; i++) {
1334 		mtcp->n_sender[i] = CreateMTCPSender(i);
1335 		if (!mtcp->n_sender[i]) {
1336 			CTRACE_ERROR("Failed to create per-nic sender structure.\n");
1337 			return NULL;
1338 		}
1339 	}
1340 
1341 	mtcp->rto_store = InitRTOHashstore();
1342 	TAILQ_INIT(&mtcp->timewait_list);
1343 	TAILQ_INIT(&mtcp->timeout_list);
1344 
1345 	return mtcp;
1346 }
1347 /*----------------------------------------------------------------------------*/
1348 static void *
1349 MTCPRunThread(void *arg)
1350 {
1351 	mctx_t mctx = (mctx_t)arg;
1352 	int cpu = mctx->cpu;
1353 	int working;
1354 	struct mtcp_manager *mtcp;
1355 	struct mtcp_thread_context *ctx;
1356 
1357 	/* affinitize the thread to this core first */
1358 	mtcp_core_affinitize(cpu);
1359 
1360 	/* memory alloc after core affinitization would use local memory
1361 	   most time */
1362 	ctx = calloc(1, sizeof(*ctx));
1363 	if (!ctx) {
1364 		perror("calloc");
1365 		TRACE_ERROR("Failed to calloc mtcp context.\n");
1366 		exit(-1);
1367 	}
1368 	ctx->thread = pthread_self();
1369 	ctx->cpu = cpu;
1370 	mtcp = ctx->mtcp_manager = InitializeMTCPManager(ctx);
1371 	if (!mtcp) {
1372 		TRACE_ERROR("Failed to initialize mtcp manager.\n");
1373 		exit(-1);
1374 	}
1375 
1376 	/* assign mtcp context's underlying I/O module */
1377 	mtcp->iom = current_iomodule_func;
1378 
1379 	/* I/O initializing */
1380 	if (mtcp->iom->init_handle)
1381 		mtcp->iom->init_handle(ctx);
1382 
1383 	if (pthread_mutex_init(&ctx->flow_pool_lock, NULL)) {
1384 		perror("pthread_mutex_init of ctx->flow_pool_lock\n");
1385 		exit(-1);
1386 	}
1387 
1388 	if (pthread_mutex_init(&ctx->socket_pool_lock, NULL)) {
1389 		perror("pthread_mutex_init of ctx->socket_pool_lock\n");
1390 		exit(-1);
1391 	}
1392 
1393 	SQ_LOCK_INIT(&ctx->connect_lock, "ctx->connect_lock", exit(-1));
1394 	SQ_LOCK_INIT(&ctx->close_lock, "ctx->close_lock", exit(-1));
1395 	SQ_LOCK_INIT(&ctx->reset_lock, "ctx->reset_lock", exit(-1));
1396 	SQ_LOCK_INIT(&ctx->sendq_lock, "ctx->sendq_lock", exit(-1));
1397 	SQ_LOCK_INIT(&ctx->ackq_lock, "ctx->ackq_lock", exit(-1));
1398 	SQ_LOCK_INIT(&ctx->destroyq_lock, "ctx->destroyq_lock", exit(-1));
1399 
1400 	/* remember this context pointer for signal processing */
1401 	g_pctx[cpu] = ctx;
1402 	mlockall(MCL_CURRENT);
1403 
1404 	// attach (nic device, queue)
1405 	working = AttachDevice(ctx);
1406 	if (working != 0) {
1407 		sem_post(&g_init_sem[ctx->cpu]);
1408 		TRACE_DBG("MTCP thread %d finished. Not attached any device\n", ctx->cpu);
1409 		pthread_exit(NULL);
1410 	}
1411 
1412 	TRACE_DBG("CPU %d: initialization finished.\n", cpu);
1413 	sem_post(&g_init_sem[ctx->cpu]);
1414 
1415 	/* start the main loop */
1416 	RunMainLoop(ctx);
1417 
1418 	TRACE_DBG("MTCP thread %d finished.\n", ctx->cpu);
1419 
1420 	/* signaling mTCP thread is done */
1421 	sem_post(&g_done_sem[mctx->cpu]);
1422 
1423 	//pthread_exit(NULL);
1424 	return 0;
1425 }
1426 /*----------------------------------------------------------------------------*/
1427 #ifdef ENABLE_DPDK
1428 static int MTCPDPDKRunThread(void *arg)
1429 {
1430 	MTCPRunThread(arg);
1431 	return 0;
1432 }
1433 #endif /* !ENABLE_DPDK */
1434 /*----------------------------------------------------------------------------*/
1435 mctx_t
1436 mtcp_create_context(int cpu)
1437 {
1438 	mctx_t mctx;
1439 	int ret;
1440 
1441 	if (cpu >=  g_config.mos->num_cores) {
1442 		TRACE_ERROR("Failed initialize new mtcp context. "
1443 					"Requested cpu id %d exceed the number of cores %d configured to use.\n",
1444 					cpu, g_config.mos->num_cores);
1445 		return NULL;
1446 	}
1447 
1448         /* check if mtcp_create_context() was already initialized */
1449         if (g_logctx[cpu] != NULL) {
1450                 TRACE_ERROR("%s was already initialized before!\n",
1451                             __FUNCTION__);
1452                 return NULL;
1453         }
1454 
1455 	ret = sem_init(&g_init_sem[cpu], 0, 0);
1456 	if (ret) {
1457 		TRACE_ERROR("Failed initialize init_sem.\n");
1458 		return NULL;
1459 	}
1460 
1461 	ret = sem_init(&g_done_sem[cpu], 0, 0);
1462 	if (ret) {
1463 		TRACE_ERROR("Failed initialize done_sem.\n");
1464 		return NULL;
1465 	}
1466 
1467 	mctx = (mctx_t)calloc(1, sizeof(struct mtcp_context));
1468 	if (!mctx) {
1469 		TRACE_ERROR("Failed to allocate memory for mtcp_context.\n");
1470 		return NULL;
1471 	}
1472 	mctx->cpu = cpu;
1473 	g_ctx[cpu] = mctx;
1474 
1475 	/* initialize logger */
1476 	g_logctx[cpu] = (struct log_thread_context *)
1477 			calloc(1, sizeof(struct log_thread_context));
1478 	if (!g_logctx[cpu]) {
1479 		perror("malloc");
1480 		TRACE_ERROR("Failed to allocate memory for log thread context.\n");
1481 		return NULL;
1482 	}
1483 	InitLogThreadContext(g_logctx[cpu], cpu);
1484 	if (pthread_create(&log_thread[cpu],
1485 			   NULL, ThreadLogMain, (void *)g_logctx[cpu])) {
1486 		perror("pthread_create");
1487 		TRACE_ERROR("Failed to create log thread\n");
1488 		return NULL;
1489 	}
1490 
1491 #ifdef ENABLE_DPDK
1492 	/* Wake up mTCP threads (wake up I/O threads) */
1493 	if (current_iomodule_func == &dpdk_module_func) {
1494 		int master;
1495 		master = rte_get_master_lcore();
1496 		if (master == cpu) {
1497 			lcore_config[master].ret = 0;
1498 			lcore_config[master].state = FINISHED;
1499 			if (pthread_create(&g_thread[cpu],
1500 					   NULL, MTCPRunThread, (void *)mctx) != 0) {
1501 				TRACE_ERROR("pthread_create of mtcp thread failed!\n");
1502 				return NULL;
1503 			}
1504 		} else
1505 			rte_eal_remote_launch(MTCPDPDKRunThread, mctx, cpu);
1506 	} else
1507 #endif /* !ENABLE_DPDK */
1508 		{
1509 			if (pthread_create(&g_thread[cpu],
1510 					   NULL, MTCPRunThread, (void *)mctx) != 0) {
1511 				TRACE_ERROR("pthread_create of mtcp thread failed!\n");
1512 				return NULL;
1513 			}
1514 		}
1515 
1516 	sem_wait(&g_init_sem[cpu]);
1517 	sem_destroy(&g_init_sem[cpu]);
1518 
1519 	running[cpu] = TRUE;
1520 
1521 #ifdef NETSTAT
1522 #if NETSTAT_TOTAL
1523 	if (printer < 0) {
1524 		printer = cpu;
1525 		TRACE_INFO("CPU %d is in charge of printing stats.\n", printer);
1526 	}
1527 #endif
1528 #endif
1529 
1530 	return mctx;
1531 }
1532 /*----------------------------------------------------------------------------*/
1533 /**
1534  * TODO: It currently always returns 0. Add appropriate error return values
1535  */
1536 int
1537 mtcp_destroy_context(mctx_t mctx)
1538 {
1539 	struct mtcp_thread_context *ctx = g_pctx[mctx->cpu];
1540 	struct mtcp_manager *mtcp = ctx->mtcp_manager;
1541 	struct log_thread_context *log_ctx = mtcp->logger;
1542 	int ret, i;
1543 
1544 	TRACE_DBG("CPU %d: mtcp_destroy_context()\n", mctx->cpu);
1545 
1546 	/* close all stream sockets that are still open */
1547 	if (!ctx->exit) {
1548 		for (i = 0; i < g_config.mos->max_concurrency; i++) {
1549 			if (mtcp->smap[i].socktype == MOS_SOCK_STREAM) {
1550 				TRACE_DBG("Closing remaining socket %d (%s)\n",
1551 						i, TCPStateToString(mtcp->smap[i].stream));
1552 #ifdef DUMP_STREAM
1553 				DumpStream(mtcp, mtcp->smap[i].stream);
1554 #endif
1555 				mtcp_close(mctx, i);
1556 			}
1557 		}
1558 	}
1559 
1560 	ctx->done = 1;
1561 
1562 	//pthread_kill(g_thread[mctx->cpu], SIGINT);
1563 #ifdef ENABLE_DPDK
1564 	ctx->exit = 1;
1565 	/* XXX - dpdk logic changes */
1566 	if (current_iomodule_func == &dpdk_module_func) {
1567 		int master = rte_get_master_lcore();
1568 		if (master == mctx->cpu)
1569 			pthread_join(g_thread[mctx->cpu], NULL);
1570 		else
1571 			rte_eal_wait_lcore(mctx->cpu);
1572 	} else
1573 #endif /* !ENABLE_DPDK */
1574 		{
1575 			pthread_join(g_thread[mctx->cpu], NULL);
1576 		}
1577 
1578 	TRACE_INFO("MTCP thread %d joined.\n", mctx->cpu);
1579 	running[mctx->cpu] = FALSE;
1580 
1581 #ifdef NETSTAT
1582 #if NETSTAT_TOTAL
1583 	if (printer == mctx->cpu) {
1584 		for (i = 0; i < num_cpus; i++) {
1585 			if (i != mctx->cpu && running[i]) {
1586 				printer = i;
1587 				break;
1588 			}
1589 		}
1590 	}
1591 #endif
1592 #endif
1593 
1594 	log_ctx->done = 1;
1595 	ret = write(log_ctx->pair_sp_fd, "F", 1);
1596 	if (ret != 1)
1597 		TRACE_ERROR("CPU %d: Fail to signal socket pair\n", mctx->cpu);
1598 
1599 	pthread_join(log_thread[ctx->cpu], NULL);
1600 	fclose(mtcp->log_fp);
1601 	TRACE_LOG("Log thread %d joined.\n", mctx->cpu);
1602 
1603 	if (mtcp->connectq) {
1604 		DestroyStreamQueue(mtcp->connectq);
1605 		mtcp->connectq = NULL;
1606 	}
1607 	if (mtcp->sendq) {
1608 		DestroyStreamQueue(mtcp->sendq);
1609 		mtcp->sendq = NULL;
1610 	}
1611 	if (mtcp->ackq) {
1612 		DestroyStreamQueue(mtcp->ackq);
1613 		mtcp->ackq = NULL;
1614 	}
1615 	if (mtcp->closeq) {
1616 		DestroyStreamQueue(mtcp->closeq);
1617 		mtcp->closeq = NULL;
1618 	}
1619 	if (mtcp->closeq_int) {
1620 		DestroyInternalStreamQueue(mtcp->closeq_int);
1621 		mtcp->closeq_int = NULL;
1622 	}
1623 	if (mtcp->resetq) {
1624 		DestroyStreamQueue(mtcp->resetq);
1625 		mtcp->resetq = NULL;
1626 	}
1627 	if (mtcp->resetq_int) {
1628 		DestroyInternalStreamQueue(mtcp->resetq_int);
1629 		mtcp->resetq_int = NULL;
1630 	}
1631 	if (mtcp->destroyq) {
1632 		DestroyStreamQueue(mtcp->destroyq);
1633 		mtcp->destroyq = NULL;
1634 	}
1635 
1636 	DestroyMTCPSender(mtcp->g_sender);
1637 	for (i = 0; i < g_config.mos->netdev_table->num; i++) {
1638 		DestroyMTCPSender(mtcp->n_sender[i]);
1639 	}
1640 
1641 	MPDestroy(mtcp->rv_pool);
1642 	MPDestroy(mtcp->sv_pool);
1643 	MPDestroy(mtcp->flow_pool);
1644 
1645 	if (mtcp->ap) {
1646 		DestroyAddressPool(mtcp->ap);
1647 	}
1648 
1649 	SQ_LOCK_DESTROY(&ctx->connect_lock);
1650 	SQ_LOCK_DESTROY(&ctx->close_lock);
1651 	SQ_LOCK_DESTROY(&ctx->reset_lock);
1652 	SQ_LOCK_DESTROY(&ctx->sendq_lock);
1653 	SQ_LOCK_DESTROY(&ctx->ackq_lock);
1654 	SQ_LOCK_DESTROY(&ctx->destroyq_lock);
1655 
1656 	//TRACE_INFO("MTCP thread %d destroyed.\n", mctx->cpu);
1657 	if (mtcp->iom->destroy_handle)
1658 		mtcp->iom->destroy_handle(ctx);
1659 	free(ctx);
1660 	free(mctx);
1661 
1662 	return 0;
1663 }
1664 /*----------------------------------------------------------------------------*/
1665 mtcp_sighandler_t
1666 mtcp_register_signal(int signum, mtcp_sighandler_t handler)
1667 {
1668 	mtcp_sighandler_t prev;
1669 
1670 	if (signum == SIGINT) {
1671 		prev = app_signal_handler;
1672 		app_signal_handler = handler;
1673 	} else {
1674 		if ((prev = signal(signum, handler)) == SIG_ERR) {
1675 			perror("signal");
1676 			return SIG_ERR;
1677 		}
1678 	}
1679 
1680 	return prev;
1681 }
1682 /*----------------------------------------------------------------------------*/
1683 int
1684 mtcp_getconf(struct mtcp_conf *conf)
1685 {
1686 	int i, j;
1687 
1688 	if (!conf)
1689 		return -1;
1690 
1691 	conf->num_cores = g_config.mos->num_cores;
1692 	conf->max_concurrency = g_config.mos->max_concurrency;
1693 	conf->cpu_mask = g_config.mos->cpu_mask;
1694 
1695 	conf->rcvbuf_size = g_config.mos->rmem_size;
1696 	conf->sndbuf_size = g_config.mos->wmem_size;
1697 
1698 	conf->tcp_timewait = g_config.mos->tcp_tw_interval;
1699 	conf->tcp_timeout = g_config.mos->tcp_timeout;
1700 
1701 	i = 0;
1702 	struct conf_block *bwalk;
1703 	TAILQ_FOREACH(bwalk, &g_config.app_blkh, link) {
1704 		struct app_conf *app_conf = (struct app_conf *)bwalk->conf;
1705 		for (j = 0; j < app_conf->app_argc; j++)
1706 			conf->app_argv[i][j] = app_conf->app_argv[j];
1707 		conf->app_argc[i] = app_conf->app_argc;
1708 		conf->app_cpu_mask[i] = app_conf->cpu_mask;
1709 		i++;
1710 	}
1711 	conf->num_app = i;
1712 
1713 	return 0;
1714 }
1715 /*----------------------------------------------------------------------------*/
1716 int
1717 mtcp_setconf(const struct mtcp_conf *conf)
1718 {
1719 	if (!conf)
1720 		return -1;
1721 
1722 	g_config.mos->num_cores = conf->num_cores;
1723 	g_config.mos->max_concurrency = conf->max_concurrency;
1724 
1725 	g_config.mos->rmem_size = conf->rcvbuf_size;
1726 	g_config.mos->wmem_size = conf->sndbuf_size;
1727 
1728 	g_config.mos->tcp_tw_interval = conf->tcp_timewait;
1729 	g_config.mos->tcp_timeout = conf->tcp_timeout;
1730 
1731 	TRACE_CONFIG("Configuration updated by mtcp_setconf().\n");
1732 	//PrintConfiguration();
1733 
1734 	return 0;
1735 }
1736 /*----------------------------------------------------------------------------*/
1737 int
1738 mtcp_init(const char *config_file)
1739 {
1740 	int i;
1741 	int ret;
1742 
1743 	if (geteuid()) {
1744 		TRACE_CONFIG("[CAUTION] Run as root if mlock is necessary.\n");
1745 #if defined(ENABLE_DPDK) || defined(ENABLE_DPDKR) || defined(ENABLE_NETMAP)
1746 		TRACE_CONFIG("[CAUTION] Run the app as root!\n");
1747 		exit(EXIT_FAILURE);
1748 #endif
1749 	}
1750 
1751 	/* getting cpu and NIC */
1752 	num_cpus = GetNumCPUs();
1753 	assert(num_cpus >= 1);
1754 	for (i = 0; i < num_cpus; i++) {
1755 		g_mtcp[i] = NULL;
1756 		running[i] = FALSE;
1757 		sigint_cnt[i] = 0;
1758 	}
1759 
1760 	ret = LoadConfigurationUpperHalf(config_file);
1761 	if (ret) {
1762 		TRACE_CONFIG("Error occured while loading configuration.\n");
1763 		return -1;
1764 	}
1765 
1766 #if defined(ENABLE_PSIO)
1767 	current_iomodule_func = &ps_module_func;
1768 #elif defined(ENABLE_DPDK)
1769 	current_iomodule_func = &dpdk_module_func;
1770 #elif defined(ENABLE_PCAP)
1771 	current_iomodule_func = &pcap_module_func;
1772 #elif defined(ENABLE_DPDKR)
1773 	current_iomodule_func = &dpdkr_module_func;
1774 #elif defined(ENABLE_NETMAP)
1775 	current_iomodule_func = &netmap_module_func;
1776 #endif
1777 
1778 	if (current_iomodule_func->load_module_upper_half)
1779 		current_iomodule_func->load_module_upper_half();
1780 
1781 	LoadConfigurationLowerHalf();
1782 
1783 	//PrintConfiguration();
1784 
1785 	for (i = 0; i < g_config.mos->netdev_table->num; i++) {
1786 		ap[i] = CreateAddressPool(g_config.mos->netdev_table->ent[i]->ip_addr, 1);
1787 		if (!ap[i]) {
1788 			TRACE_CONFIG("Error occured while create address pool[%d]\n",
1789 				     i);
1790 			return -1;
1791 		}
1792 	}
1793 
1794 	//PrintInterfaceInfo();
1795 	//PrintRoutingTable();
1796 	//PrintARPTable();
1797 	InitARPTable();
1798 
1799 	if (signal(SIGUSR1, HandleSignal) == SIG_ERR) {
1800 		perror("signal, SIGUSR1");
1801 		return -1;
1802 	}
1803 	if (signal(SIGINT, HandleSignal) == SIG_ERR) {
1804 		perror("signal, SIGINT");
1805 		return -1;
1806 	}
1807 	app_signal_handler = NULL;
1808 
1809 	printf("load_module(): %p\n", current_iomodule_func);
1810 	/* load system-wide io module specs */
1811 	if (current_iomodule_func->load_module_lower_half)
1812 		current_iomodule_func->load_module_lower_half();
1813 
1814 	GlobInitEvent();
1815 
1816 	PrintConf(&g_config);
1817 
1818 	return 0;
1819 }
1820 /*----------------------------------------------------------------------------*/
1821 int
1822 mtcp_destroy()
1823 {
1824 	int i;
1825 
1826 	/* wait until all threads are closed */
1827 	for (i = 0; i < num_cpus; i++) {
1828 		if (running[i]) {
1829 			if (pthread_join(g_thread[i], NULL) != 0)
1830 				return -1;
1831 		}
1832 	}
1833 
1834 	for (i = 0; i < g_config.mos->netdev_table->num; i++)
1835 		DestroyAddressPool(ap[i]);
1836 
1837 	TRACE_INFO("All MTCP threads are joined.\n");
1838 
1839 	return 0;
1840 }
1841 /*----------------------------------------------------------------------------*/
1842