• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 #include <stdio.h>
2 #include <stdlib.h>
3 #include <unistd.h>
4 #include <limits.h>
5 #include <errno.h>
6 #include <fcntl.h>
7 #include <sys/poll.h>
8 #include <sys/types.h>
9 #include <sys/stat.h>
10 #include <sys/wait.h>
11 #include <sys/socket.h>
12 #include <sys/un.h>
13 #include <netinet/in.h>
14 #include <arpa/inet.h>
15 #include <netdb.h>
16 #include <signal.h>
17 #ifdef CONFIG_ZLIB
18 #include <zlib.h>
19 #endif
20 
21 #include "fio.h"
22 #include "client.h"
23 #include "server.h"
24 #include "flist.h"
25 #include "hash.h"
26 #include "verify.h"
27 
28 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
29 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd);
30 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd);
31 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
32 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd);
33 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd);
34 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd);
35 
36 static void convert_text(struct fio_net_cmd *cmd);
37 
38 struct client_ops fio_client_ops = {
39 	.text		= handle_text,
40 	.disk_util	= handle_du,
41 	.thread_status	= handle_ts,
42 	.group_stats	= handle_gs,
43 	.stop		= handle_stop,
44 	.start		= handle_start,
45 	.eta		= display_thread_status,
46 	.probe		= handle_probe,
47 	.eta_msec	= FIO_CLIENT_DEF_ETA_MSEC,
48 	.client_type	= FIO_CLIENT_TYPE_CLI,
49 };
50 
51 static struct timeval eta_tv;
52 
53 static FLIST_HEAD(client_list);
54 static FLIST_HEAD(eta_list);
55 
56 static FLIST_HEAD(arg_list);
57 
58 struct thread_stat client_ts;
59 struct group_run_stats client_gs;
60 int sum_stat_clients;
61 
62 static int sum_stat_nr;
63 static struct json_object *root = NULL;
64 static struct json_object *job_opt_object = NULL;
65 static struct json_array *clients_array = NULL;
66 static struct json_array *du_array = NULL;
67 
68 static int error_clients;
69 
70 #define FIO_CLIENT_HASH_BITS	7
71 #define FIO_CLIENT_HASH_SZ	(1 << FIO_CLIENT_HASH_BITS)
72 #define FIO_CLIENT_HASH_MASK	(FIO_CLIENT_HASH_SZ - 1)
73 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
74 
75 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *, bool *);
76 
fio_client_add_hash(struct fio_client * client)77 static void fio_client_add_hash(struct fio_client *client)
78 {
79 	int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
80 
81 	bucket &= FIO_CLIENT_HASH_MASK;
82 	flist_add(&client->hash_list, &client_hash[bucket]);
83 }
84 
fio_client_remove_hash(struct fio_client * client)85 static void fio_client_remove_hash(struct fio_client *client)
86 {
87 	if (!flist_empty(&client->hash_list))
88 		flist_del_init(&client->hash_list);
89 }
90 
fio_client_hash_init(void)91 static void fio_init fio_client_hash_init(void)
92 {
93 	int i;
94 
95 	for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
96 		INIT_FLIST_HEAD(&client_hash[i]);
97 }
98 
read_data(int fd,void * data,size_t size)99 static int read_data(int fd, void *data, size_t size)
100 {
101 	ssize_t ret;
102 
103 	while (size) {
104 		ret = read(fd, data, size);
105 		if (ret < 0) {
106 			if (errno == EAGAIN || errno == EINTR)
107 				continue;
108 			break;
109 		} else if (!ret)
110 			break;
111 		else {
112 			data += ret;
113 			size -= ret;
114 		}
115 	}
116 
117 	if (size)
118 		return EAGAIN;
119 
120 	return 0;
121 }
122 
fio_client_json_init(void)123 static void fio_client_json_init(void)
124 {
125 	char time_buf[32];
126 	time_t time_p;
127 
128 	if (!(output_format & FIO_OUTPUT_JSON))
129 		return;
130 
131 	time(&time_p);
132 	os_ctime_r((const time_t *) &time_p, time_buf, sizeof(time_buf));
133 	time_buf[strlen(time_buf) - 1] = '\0';
134 
135 	root = json_create_object();
136 	json_object_add_value_string(root, "fio version", fio_version_string);
137 	json_object_add_value_int(root, "timestamp", time_p);
138 	json_object_add_value_string(root, "time", time_buf);
139 
140 	job_opt_object = json_create_object();
141 	json_object_add_value_object(root, "global options", job_opt_object);
142 	clients_array = json_create_array();
143 	json_object_add_value_array(root, "client_stats", clients_array);
144 	du_array = json_create_array();
145 	json_object_add_value_array(root, "disk_util", du_array);
146 }
147 
fio_client_json_fini(void)148 static void fio_client_json_fini(void)
149 {
150 	if (!(output_format & FIO_OUTPUT_JSON))
151 		return;
152 
153 	log_info("\n");
154 	json_print_object(root, NULL);
155 	log_info("\n");
156 	json_free_object(root);
157 	root = NULL;
158 	clients_array = NULL;
159 	du_array = NULL;
160 }
161 
find_client_by_fd(int fd)162 static struct fio_client *find_client_by_fd(int fd)
163 {
164 	int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
165 	struct fio_client *client;
166 	struct flist_head *entry;
167 
168 	flist_for_each(entry, &client_hash[bucket]) {
169 		client = flist_entry(entry, struct fio_client, hash_list);
170 
171 		if (client->fd == fd) {
172 			client->refs++;
173 			return client;
174 		}
175 	}
176 
177 	return NULL;
178 }
179 
fio_put_client(struct fio_client * client)180 void fio_put_client(struct fio_client *client)
181 {
182 	if (--client->refs)
183 		return;
184 
185 	free(client->hostname);
186 	if (client->argv)
187 		free(client->argv);
188 	if (client->name)
189 		free(client->name);
190 	while (client->nr_files) {
191 		struct client_file *cf = &client->files[--client->nr_files];
192 
193 		free(cf->file);
194 	}
195 	if (client->files)
196 		free(client->files);
197 	if (client->opt_lists)
198 		free(client->opt_lists);
199 
200 	if (!client->did_stat)
201 		sum_stat_clients--;
202 
203 	if (client->error)
204 		error_clients++;
205 
206 	free(client);
207 }
208 
fio_client_dec_jobs_eta(struct client_eta * eta,client_eta_op eta_fn)209 static int fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
210 {
211 	if (!--eta->pending) {
212 		eta_fn(&eta->eta);
213 		free(eta);
214 		return 0;
215 	}
216 
217 	return 1;
218 }
219 
fio_drain_client_text(struct fio_client * client)220 static void fio_drain_client_text(struct fio_client *client)
221 {
222 	do {
223 		struct fio_net_cmd *cmd;
224 
225 		cmd = fio_net_recv_cmd(client->fd, false);
226 		if (!cmd)
227 			break;
228 
229 		if (cmd->opcode == FIO_NET_CMD_TEXT) {
230 			convert_text(cmd);
231 			client->ops->text(client, cmd);
232 		}
233 
234 		free(cmd);
235 	} while (1);
236 }
237 
remove_client(struct fio_client * client)238 static void remove_client(struct fio_client *client)
239 {
240 	assert(client->refs);
241 
242 	dprint(FD_NET, "client: removed <%s>\n", client->hostname);
243 
244 	fio_drain_client_text(client);
245 
246 	if (!flist_empty(&client->list))
247 		flist_del_init(&client->list);
248 
249 	fio_client_remove_hash(client);
250 
251 	if (!flist_empty(&client->eta_list)) {
252 		flist_del_init(&client->eta_list);
253 		fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
254 	}
255 
256 	close(client->fd);
257 	client->fd = -1;
258 
259 	if (client->ops->removed)
260 		client->ops->removed(client);
261 
262 	nr_clients--;
263 	fio_put_client(client);
264 }
265 
fio_get_client(struct fio_client * client)266 struct fio_client *fio_get_client(struct fio_client *client)
267 {
268 	client->refs++;
269 	return client;
270 }
271 
__fio_client_add_cmd_option(struct fio_client * client,const char * opt)272 static void __fio_client_add_cmd_option(struct fio_client *client,
273 					const char *opt)
274 {
275 	int index;
276 
277 	index = client->argc++;
278 	client->argv = realloc(client->argv, sizeof(char *) * client->argc);
279 	client->argv[index] = strdup(opt);
280 	dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
281 }
282 
fio_client_add_cmd_option(void * cookie,const char * opt)283 void fio_client_add_cmd_option(void *cookie, const char *opt)
284 {
285 	struct fio_client *client = cookie;
286 	struct flist_head *entry;
287 
288 	if (!client || !opt)
289 		return;
290 
291 	__fio_client_add_cmd_option(client, opt);
292 
293 	/*
294 	 * Duplicate arguments to shared client group
295 	 */
296 	flist_for_each(entry, &arg_list) {
297 		client = flist_entry(entry, struct fio_client, arg_list);
298 
299 		__fio_client_add_cmd_option(client, opt);
300 	}
301 }
302 
fio_client_add_explicit(struct client_ops * ops,const char * hostname,int type,int port)303 struct fio_client *fio_client_add_explicit(struct client_ops *ops,
304 					   const char *hostname, int type,
305 					   int port)
306 {
307 	struct fio_client *client;
308 
309 	client = malloc(sizeof(*client));
310 	memset(client, 0, sizeof(*client));
311 
312 	INIT_FLIST_HEAD(&client->list);
313 	INIT_FLIST_HEAD(&client->hash_list);
314 	INIT_FLIST_HEAD(&client->arg_list);
315 	INIT_FLIST_HEAD(&client->eta_list);
316 	INIT_FLIST_HEAD(&client->cmd_list);
317 
318 	client->hostname = strdup(hostname);
319 
320 	if (type == Fio_client_socket)
321 		client->is_sock = 1;
322 	else {
323 		int ipv6;
324 
325 		ipv6 = type == Fio_client_ipv6;
326 		if (fio_server_parse_host(hostname, ipv6,
327 						&client->addr.sin_addr,
328 						&client->addr6.sin6_addr))
329 			goto err;
330 
331 		client->port = port;
332 	}
333 
334 	client->fd = -1;
335 	client->ops = ops;
336 	client->refs = 1;
337 	client->type = ops->client_type;
338 
339 	__fio_client_add_cmd_option(client, "fio");
340 
341 	flist_add(&client->list, &client_list);
342 	nr_clients++;
343 	dprint(FD_NET, "client: added <%s>\n", client->hostname);
344 	return client;
345 err:
346 	free(client);
347 	return NULL;
348 }
349 
fio_client_add_ini_file(void * cookie,const char * ini_file,bool remote)350 int fio_client_add_ini_file(void *cookie, const char *ini_file, bool remote)
351 {
352 	struct fio_client *client = cookie;
353 	struct client_file *cf;
354 	size_t new_size;
355 	void *new_files;
356 
357 	if (!client)
358 		return 1;
359 
360 	dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file);
361 
362 	new_size = (client->nr_files + 1) * sizeof(struct client_file);
363 	new_files = realloc(client->files, new_size);
364 	if (!new_files)
365 		return 1;
366 
367 	client->files = new_files;
368 	cf = &client->files[client->nr_files];
369 	cf->file = strdup(ini_file);
370 	cf->remote = remote;
371 	client->nr_files++;
372 	return 0;
373 }
374 
fio_client_add(struct client_ops * ops,const char * hostname,void ** cookie)375 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
376 {
377 	struct fio_client *existing = *cookie;
378 	struct fio_client *client;
379 
380 	if (existing) {
381 		/*
382 		 * We always add our "exec" name as the option, hence 1
383 		 * means empty.
384 		 */
385 		if (existing->argc == 1)
386 			flist_add_tail(&existing->arg_list, &arg_list);
387 		else {
388 			while (!flist_empty(&arg_list))
389 				flist_del_init(arg_list.next);
390 		}
391 	}
392 
393 	client = malloc(sizeof(*client));
394 	memset(client, 0, sizeof(*client));
395 
396 	INIT_FLIST_HEAD(&client->list);
397 	INIT_FLIST_HEAD(&client->hash_list);
398 	INIT_FLIST_HEAD(&client->arg_list);
399 	INIT_FLIST_HEAD(&client->eta_list);
400 	INIT_FLIST_HEAD(&client->cmd_list);
401 
402 	if (fio_server_parse_string(hostname, &client->hostname,
403 					&client->is_sock, &client->port,
404 					&client->addr.sin_addr,
405 					&client->addr6.sin6_addr,
406 					&client->ipv6))
407 		return -1;
408 
409 	client->fd = -1;
410 	client->ops = ops;
411 	client->refs = 1;
412 	client->type = ops->client_type;
413 
414 	__fio_client_add_cmd_option(client, "fio");
415 
416 	flist_add(&client->list, &client_list);
417 	nr_clients++;
418 	dprint(FD_NET, "client: added <%s>\n", client->hostname);
419 	*cookie = client;
420 	return 0;
421 }
422 
server_name(struct fio_client * client,char * buf,size_t bufsize)423 static const char *server_name(struct fio_client *client, char *buf,
424 			       size_t bufsize)
425 {
426 	const char *from;
427 
428 	if (client->ipv6)
429 		from = inet_ntop(AF_INET6, (struct sockaddr *) &client->addr6.sin6_addr, buf, bufsize);
430 	else if (client->is_sock)
431 		from = "sock";
432 	else
433 		from = inet_ntop(AF_INET, (struct sockaddr *) &client->addr.sin_addr, buf, bufsize);
434 
435 	return from;
436 }
437 
probe_client(struct fio_client * client)438 static void probe_client(struct fio_client *client)
439 {
440 	struct cmd_client_probe_pdu pdu;
441 	const char *sname;
442 	uint64_t tag;
443 	char buf[64];
444 
445 	dprint(FD_NET, "client: send probe\n");
446 
447 #ifdef CONFIG_ZLIB
448 	pdu.flags = __le64_to_cpu(FIO_PROBE_FLAG_ZLIB);
449 #else
450 	pdu.flags = 0;
451 #endif
452 
453 	sname = server_name(client, buf, sizeof(buf));
454 	memset(pdu.server, 0, sizeof(pdu.server));
455 	strncpy((char *) pdu.server, sname, sizeof(pdu.server) - 1);
456 
457 	fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list);
458 }
459 
fio_client_connect_ip(struct fio_client * client)460 static int fio_client_connect_ip(struct fio_client *client)
461 {
462 	struct sockaddr *addr;
463 	socklen_t socklen;
464 	int fd, domain;
465 
466 	if (client->ipv6) {
467 		client->addr6.sin6_family = AF_INET6;
468 		client->addr6.sin6_port = htons(client->port);
469 		domain = AF_INET6;
470 		addr = (struct sockaddr *) &client->addr6;
471 		socklen = sizeof(client->addr6);
472 	} else {
473 		client->addr.sin_family = AF_INET;
474 		client->addr.sin_port = htons(client->port);
475 		domain = AF_INET;
476 		addr = (struct sockaddr *) &client->addr;
477 		socklen = sizeof(client->addr);
478 	}
479 
480 	fd = socket(domain, SOCK_STREAM, 0);
481 	if (fd < 0) {
482 		int ret = -errno;
483 
484 		log_err("fio: socket: %s\n", strerror(errno));
485 		return ret;
486 	}
487 
488 	if (connect(fd, addr, socklen) < 0) {
489 		int ret = -errno;
490 
491 		log_err("fio: connect: %s\n", strerror(errno));
492 		log_err("fio: failed to connect to %s:%u\n", client->hostname,
493 								client->port);
494 		close(fd);
495 		return ret;
496 	}
497 
498 	return fd;
499 }
500 
fio_client_connect_sock(struct fio_client * client)501 static int fio_client_connect_sock(struct fio_client *client)
502 {
503 	struct sockaddr_un *addr = &client->addr_un;
504 	socklen_t len;
505 	int fd;
506 
507 	memset(addr, 0, sizeof(*addr));
508 	addr->sun_family = AF_UNIX;
509 	strncpy(addr->sun_path, client->hostname, sizeof(addr->sun_path) - 1);
510 
511 	fd = socket(AF_UNIX, SOCK_STREAM, 0);
512 	if (fd < 0) {
513 		int ret = -errno;
514 
515 		log_err("fio: socket: %s\n", strerror(errno));
516 		return ret;
517 	}
518 
519 	len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
520 	if (connect(fd, (struct sockaddr *) addr, len) < 0) {
521 		int ret = -errno;
522 
523 		log_err("fio: connect; %s\n", strerror(errno));
524 		close(fd);
525 		return ret;
526 	}
527 
528 	return fd;
529 }
530 
fio_client_connect(struct fio_client * client)531 int fio_client_connect(struct fio_client *client)
532 {
533 	int fd;
534 
535 	dprint(FD_NET, "client: connect to host %s\n", client->hostname);
536 
537 	if (client->is_sock)
538 		fd = fio_client_connect_sock(client);
539 	else
540 		fd = fio_client_connect_ip(client);
541 
542 	dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
543 
544 	if (fd < 0)
545 		return fd;
546 
547 	client->fd = fd;
548 	fio_client_add_hash(client);
549 	client->state = Client_connected;
550 
551 	probe_client(client);
552 	return 0;
553 }
554 
fio_client_terminate(struct fio_client * client)555 int fio_client_terminate(struct fio_client *client)
556 {
557 	return fio_net_send_quit(client->fd);
558 }
559 
fio_clients_terminate(void)560 static void fio_clients_terminate(void)
561 {
562 	struct flist_head *entry;
563 	struct fio_client *client;
564 
565 	dprint(FD_NET, "client: terminate clients\n");
566 
567 	flist_for_each(entry, &client_list) {
568 		client = flist_entry(entry, struct fio_client, list);
569 		fio_client_terminate(client);
570 	}
571 }
572 
sig_int(int sig)573 static void sig_int(int sig)
574 {
575 	dprint(FD_NET, "client: got signal %d\n", sig);
576 	fio_clients_terminate();
577 }
578 
client_signal_handler(void)579 static void client_signal_handler(void)
580 {
581 	struct sigaction act;
582 
583 	memset(&act, 0, sizeof(act));
584 	act.sa_handler = sig_int;
585 	act.sa_flags = SA_RESTART;
586 	sigaction(SIGINT, &act, NULL);
587 
588 	memset(&act, 0, sizeof(act));
589 	act.sa_handler = sig_int;
590 	act.sa_flags = SA_RESTART;
591 	sigaction(SIGTERM, &act, NULL);
592 
593 /* Windows uses SIGBREAK as a quit signal from other applications */
594 #ifdef WIN32
595 	memset(&act, 0, sizeof(act));
596 	act.sa_handler = sig_int;
597 	act.sa_flags = SA_RESTART;
598 	sigaction(SIGBREAK, &act, NULL);
599 #endif
600 
601 	memset(&act, 0, sizeof(act));
602 	act.sa_handler = sig_show_status;
603 	act.sa_flags = SA_RESTART;
604 	sigaction(SIGUSR1, &act, NULL);
605 }
606 
send_client_cmd_line(struct fio_client * client)607 static int send_client_cmd_line(struct fio_client *client)
608 {
609 	struct cmd_single_line_pdu *cslp;
610 	struct cmd_line_pdu *clp;
611 	unsigned long offset;
612 	unsigned int *lens;
613 	void *pdu;
614 	size_t mem;
615 	int i, ret;
616 
617 	dprint(FD_NET, "client: send cmdline %d\n", client->argc);
618 
619 	lens = malloc(client->argc * sizeof(unsigned int));
620 
621 	/*
622 	 * Find out how much mem we need
623 	 */
624 	for (i = 0, mem = 0; i < client->argc; i++) {
625 		lens[i] = strlen(client->argv[i]) + 1;
626 		mem += lens[i];
627 	}
628 
629 	/*
630 	 * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
631 	 */
632 	mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
633 
634 	pdu = malloc(mem);
635 	clp = pdu;
636 	offset = sizeof(*clp);
637 
638 	for (i = 0; i < client->argc; i++) {
639 		uint16_t arg_len = lens[i];
640 
641 		cslp = pdu + offset;
642 		strcpy((char *) cslp->text, client->argv[i]);
643 		cslp->len = cpu_to_le16(arg_len);
644 		offset += sizeof(*cslp) + arg_len;
645 	}
646 
647 	free(lens);
648 	clp->lines = cpu_to_le16(client->argc);
649 	clp->client_type = __cpu_to_le16(client->type);
650 	ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, NULL, NULL);
651 	free(pdu);
652 	return ret;
653 }
654 
fio_clients_connect(void)655 int fio_clients_connect(void)
656 {
657 	struct fio_client *client;
658 	struct flist_head *entry, *tmp;
659 	int ret;
660 
661 #ifdef WIN32
662 	WSADATA wsd;
663 	WSAStartup(MAKEWORD(2, 2), &wsd);
664 #endif
665 
666 	dprint(FD_NET, "client: connect all\n");
667 
668 	client_signal_handler();
669 
670 	flist_for_each_safe(entry, tmp, &client_list) {
671 		client = flist_entry(entry, struct fio_client, list);
672 
673 		ret = fio_client_connect(client);
674 		if (ret) {
675 			remove_client(client);
676 			continue;
677 		}
678 
679 		if (client->argc > 1)
680 			send_client_cmd_line(client);
681 	}
682 
683 	return !nr_clients;
684 }
685 
fio_start_client(struct fio_client * client)686 int fio_start_client(struct fio_client *client)
687 {
688 	dprint(FD_NET, "client: start %s\n", client->hostname);
689 	return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
690 }
691 
fio_start_all_clients(void)692 int fio_start_all_clients(void)
693 {
694 	struct fio_client *client;
695 	struct flist_head *entry, *tmp;
696 	int ret;
697 
698 	dprint(FD_NET, "client: start all\n");
699 
700 	fio_client_json_init();
701 
702 	flist_for_each_safe(entry, tmp, &client_list) {
703 		client = flist_entry(entry, struct fio_client, list);
704 
705 		ret = fio_start_client(client);
706 		if (ret) {
707 			remove_client(client);
708 			continue;
709 		}
710 	}
711 
712 	return flist_empty(&client_list);
713 }
714 
__fio_client_send_remote_ini(struct fio_client * client,const char * filename)715 static int __fio_client_send_remote_ini(struct fio_client *client,
716 					const char *filename)
717 {
718 	struct cmd_load_file_pdu *pdu;
719 	size_t p_size;
720 	int ret;
721 
722 	dprint(FD_NET, "send remote ini %s to %s\n", filename, client->hostname);
723 
724 	p_size = sizeof(*pdu) + strlen(filename) + 1;
725 	pdu = malloc(p_size);
726 	memset(pdu, 0, p_size);
727 	pdu->name_len = strlen(filename);
728 	strcpy((char *) pdu->file, filename);
729 	pdu->client_type = cpu_to_le16((uint16_t) client->type);
730 
731 	client->sent_job = 1;
732 	ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_LOAD_FILE, pdu, p_size,NULL, NULL);
733 	free(pdu);
734 	return ret;
735 }
736 
737 /*
738  * Send file contents to server backend. We could use sendfile(), but to remain
739  * more portable lets just read/write the darn thing.
740  */
__fio_client_send_local_ini(struct fio_client * client,const char * filename)741 static int __fio_client_send_local_ini(struct fio_client *client,
742 				       const char *filename)
743 {
744 	struct cmd_job_pdu *pdu;
745 	size_t p_size;
746 	struct stat sb;
747 	char *p;
748 	void *buf;
749 	off_t len;
750 	int fd, ret;
751 
752 	dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
753 
754 	fd = open(filename, O_RDONLY);
755 	if (fd < 0) {
756 		ret = -errno;
757 		log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
758 		return ret;
759 	}
760 
761 	if (fstat(fd, &sb) < 0) {
762 		ret = -errno;
763 		log_err("fio: job file stat: %s\n", strerror(errno));
764 		close(fd);
765 		return ret;
766 	}
767 
768 	p_size = sb.st_size + sizeof(*pdu);
769 	pdu = malloc(p_size);
770 	buf = pdu->buf;
771 
772 	len = sb.st_size;
773 	p = buf;
774 	if (read_data(fd, p, len)) {
775 		log_err("fio: failed reading job file %s\n", filename);
776 		close(fd);
777 		free(pdu);
778 		return 1;
779 	}
780 
781 	pdu->buf_len = __cpu_to_le32(sb.st_size);
782 	pdu->client_type = cpu_to_le32(client->type);
783 
784 	client->sent_job = 1;
785 	ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, NULL, NULL);
786 	free(pdu);
787 	close(fd);
788 	return ret;
789 }
790 
fio_client_send_ini(struct fio_client * client,const char * filename,bool remote)791 int fio_client_send_ini(struct fio_client *client, const char *filename,
792 			bool remote)
793 {
794 	int ret;
795 
796 	if (!remote)
797 		ret = __fio_client_send_local_ini(client, filename);
798 	else
799 		ret = __fio_client_send_remote_ini(client, filename);
800 
801 	if (!ret)
802 		client->sent_job = 1;
803 
804 	return ret;
805 }
806 
fio_client_send_cf(struct fio_client * client,struct client_file * cf)807 static int fio_client_send_cf(struct fio_client *client,
808 			      struct client_file *cf)
809 {
810 	return fio_client_send_ini(client, cf->file, cf->remote);
811 }
812 
fio_clients_send_ini(const char * filename)813 int fio_clients_send_ini(const char *filename)
814 {
815 	struct fio_client *client;
816 	struct flist_head *entry, *tmp;
817 
818 	flist_for_each_safe(entry, tmp, &client_list) {
819 		bool failed = false;
820 
821 		client = flist_entry(entry, struct fio_client, list);
822 
823 		if (client->nr_files) {
824 			int i;
825 
826 			for (i = 0; i < client->nr_files; i++) {
827 				struct client_file *cf;
828 
829 				cf = &client->files[i];
830 
831 				if (fio_client_send_cf(client, cf)) {
832 					failed = true;
833 					remove_client(client);
834 					break;
835 				}
836 			}
837 		}
838 		if (client->sent_job || failed)
839 			continue;
840 		if (!filename || fio_client_send_ini(client, filename, 0))
841 			remove_client(client);
842 	}
843 
844 	return !nr_clients;
845 }
846 
fio_client_update_options(struct fio_client * client,struct thread_options * o,uint64_t * tag)847 int fio_client_update_options(struct fio_client *client,
848 			      struct thread_options *o, uint64_t *tag)
849 {
850 	struct cmd_add_job_pdu pdu;
851 
852 	pdu.thread_number = cpu_to_le32(client->thread_number);
853 	pdu.groupid = cpu_to_le32(client->groupid);
854 	convert_thread_options_to_net(&pdu.top, o);
855 
856 	return fio_net_send_cmd(client->fd, FIO_NET_CMD_UPDATE_JOB, &pdu, sizeof(pdu), tag, &client->cmd_list);
857 }
858 
convert_io_stat(struct io_stat * dst,struct io_stat * src)859 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
860 {
861 	dst->max_val	= le64_to_cpu(src->max_val);
862 	dst->min_val	= le64_to_cpu(src->min_val);
863 	dst->samples	= le64_to_cpu(src->samples);
864 
865 	/*
866 	 * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
867 	 */
868 	dst->mean.u.f	= fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
869 	dst->S.u.f	= fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
870 }
871 
convert_ts(struct thread_stat * dst,struct thread_stat * src)872 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
873 {
874 	int i, j;
875 
876 	dst->error		= le32_to_cpu(src->error);
877 	dst->thread_number	= le32_to_cpu(src->thread_number);
878 	dst->groupid		= le32_to_cpu(src->groupid);
879 	dst->pid		= le32_to_cpu(src->pid);
880 	dst->members		= le32_to_cpu(src->members);
881 	dst->unified_rw_rep	= le32_to_cpu(src->unified_rw_rep);
882 
883 	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
884 		convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
885 		convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
886 		convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
887 		convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
888 	}
889 
890 	dst->usr_time		= le64_to_cpu(src->usr_time);
891 	dst->sys_time		= le64_to_cpu(src->sys_time);
892 	dst->ctx		= le64_to_cpu(src->ctx);
893 	dst->minf		= le64_to_cpu(src->minf);
894 	dst->majf		= le64_to_cpu(src->majf);
895 	dst->clat_percentiles	= le64_to_cpu(src->clat_percentiles);
896 	dst->percentile_precision = le64_to_cpu(src->percentile_precision);
897 
898 	for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
899 		fio_fp64_t *fps = &src->percentile_list[i];
900 		fio_fp64_t *fpd = &dst->percentile_list[i];
901 
902 		fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
903 	}
904 
905 	for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
906 		dst->io_u_map[i]	= le32_to_cpu(src->io_u_map[i]);
907 		dst->io_u_submit[i]	= le32_to_cpu(src->io_u_submit[i]);
908 		dst->io_u_complete[i]	= le32_to_cpu(src->io_u_complete[i]);
909 	}
910 
911 	for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
912 		dst->io_u_lat_u[i]	= le32_to_cpu(src->io_u_lat_u[i]);
913 	for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
914 		dst->io_u_lat_m[i]	= le32_to_cpu(src->io_u_lat_m[i]);
915 
916 	for (i = 0; i < DDIR_RWDIR_CNT; i++)
917 		for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
918 			dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
919 
920 	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
921 		dst->total_io_u[i]	= le64_to_cpu(src->total_io_u[i]);
922 		dst->short_io_u[i]	= le64_to_cpu(src->short_io_u[i]);
923 		dst->drop_io_u[i]	= le64_to_cpu(src->drop_io_u[i]);
924 	}
925 
926 	dst->total_submit	= le64_to_cpu(src->total_submit);
927 	dst->total_complete	= le64_to_cpu(src->total_complete);
928 
929 	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
930 		dst->io_bytes[i]	= le64_to_cpu(src->io_bytes[i]);
931 		dst->runtime[i]		= le64_to_cpu(src->runtime[i]);
932 	}
933 
934 	dst->total_run_time	= le64_to_cpu(src->total_run_time);
935 	dst->continue_on_error	= le16_to_cpu(src->continue_on_error);
936 	dst->total_err_count	= le64_to_cpu(src->total_err_count);
937 	dst->first_error	= le32_to_cpu(src->first_error);
938 	dst->kb_base		= le32_to_cpu(src->kb_base);
939 	dst->unit_base		= le32_to_cpu(src->unit_base);
940 
941 	dst->latency_depth	= le32_to_cpu(src->latency_depth);
942 	dst->latency_target	= le64_to_cpu(src->latency_target);
943 	dst->latency_window	= le64_to_cpu(src->latency_window);
944 	dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
945 
946 	dst->nr_block_infos	= le64_to_cpu(src->nr_block_infos);
947 	for (i = 0; i < dst->nr_block_infos; i++)
948 		dst->block_infos[i] = le32_to_cpu(src->block_infos[i]);
949 
950 	dst->ss_dur		= le64_to_cpu(src->ss_dur);
951 	dst->ss_state		= le32_to_cpu(src->ss_state);
952 	dst->ss_head		= le32_to_cpu(src->ss_head);
953 	dst->ss_limit.u.f 	= fio_uint64_to_double(le64_to_cpu(src->ss_limit.u.i));
954 	dst->ss_slope.u.f 	= fio_uint64_to_double(le64_to_cpu(src->ss_slope.u.i));
955 	dst->ss_deviation.u.f 	= fio_uint64_to_double(le64_to_cpu(src->ss_deviation.u.i));
956 	dst->ss_criterion.u.f 	= fio_uint64_to_double(le64_to_cpu(src->ss_criterion.u.i));
957 
958 	if (dst->ss_state & __FIO_SS_DATA) {
959 		for (i = 0; i < dst->ss_dur; i++ ) {
960 			dst->ss_iops_data[i] = le64_to_cpu(src->ss_iops_data[i]);
961 			dst->ss_bw_data[i] = le64_to_cpu(src->ss_bw_data[i]);
962 		}
963 	}
964 }
965 
convert_gs(struct group_run_stats * dst,struct group_run_stats * src)966 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
967 {
968 	int i;
969 
970 	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
971 		dst->max_run[i]		= le64_to_cpu(src->max_run[i]);
972 		dst->min_run[i]		= le64_to_cpu(src->min_run[i]);
973 		dst->max_bw[i]		= le64_to_cpu(src->max_bw[i]);
974 		dst->min_bw[i]		= le64_to_cpu(src->min_bw[i]);
975 		dst->iobytes[i]		= le64_to_cpu(src->iobytes[i]);
976 		dst->agg[i]		= le64_to_cpu(src->agg[i]);
977 	}
978 
979 	dst->kb_base	= le32_to_cpu(src->kb_base);
980 	dst->unit_base	= le32_to_cpu(src->unit_base);
981 	dst->groupid	= le32_to_cpu(src->groupid);
982 	dst->unified_rw_rep	= le32_to_cpu(src->unified_rw_rep);
983 }
984 
json_object_add_client_info(struct json_object * obj,struct fio_client * client)985 static void json_object_add_client_info(struct json_object *obj,
986 					struct fio_client *client)
987 {
988 	const char *hostname = client->hostname ? client->hostname : "";
989 
990 	json_object_add_value_string(obj, "hostname", hostname);
991 	json_object_add_value_int(obj, "port", client->port);
992 }
993 
handle_ts(struct fio_client * client,struct fio_net_cmd * cmd)994 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
995 {
996 	struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
997 	struct flist_head *opt_list = NULL;
998 	struct json_object *tsobj;
999 
1000 	if (client->opt_lists && p->ts.thread_number <= client->jobs)
1001 		opt_list = &client->opt_lists[p->ts.thread_number - 1];
1002 
1003 	tsobj = show_thread_status(&p->ts, &p->rs, opt_list, NULL);
1004 	client->did_stat = 1;
1005 	if (tsobj) {
1006 		json_object_add_client_info(tsobj, client);
1007 		json_array_add_value_object(clients_array, tsobj);
1008 	}
1009 
1010 	if (sum_stat_clients <= 1)
1011 		return;
1012 
1013 	sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1);
1014 	sum_group_stats(&client_gs, &p->rs);
1015 
1016 	client_ts.members++;
1017 	client_ts.thread_number = p->ts.thread_number;
1018 	client_ts.groupid = p->ts.groupid;
1019 	client_ts.unified_rw_rep = p->ts.unified_rw_rep;
1020 
1021 	if (++sum_stat_nr == sum_stat_clients) {
1022 		strcpy(client_ts.name, "All clients");
1023 		tsobj = show_thread_status(&client_ts, &client_gs, NULL, NULL);
1024 		if (tsobj) {
1025 			json_object_add_client_info(tsobj, client);
1026 			json_array_add_value_object(clients_array, tsobj);
1027 		}
1028 	}
1029 }
1030 
handle_gs(struct fio_client * client,struct fio_net_cmd * cmd)1031 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
1032 {
1033 	struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1034 
1035 	if (output_format & FIO_OUTPUT_NORMAL)
1036 		show_group_stats(gs, NULL);
1037 }
1038 
handle_job_opt(struct fio_client * client,struct fio_net_cmd * cmd)1039 static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
1040 {
1041 	struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
1042 	struct print_option *p;
1043 
1044 	if (!job_opt_object)
1045 		return;
1046 
1047 	pdu->global = le16_to_cpu(pdu->global);
1048 	pdu->truncated = le16_to_cpu(pdu->truncated);
1049 	pdu->groupid = le32_to_cpu(pdu->groupid);
1050 
1051 	p = malloc(sizeof(*p));
1052 	p->name = strdup((char *) pdu->name);
1053 	if (pdu->value[0] != '\0')
1054 		p->value = strdup((char *) pdu->value);
1055 	else
1056 		p->value = NULL;
1057 
1058 	if (pdu->global) {
1059 		const char *pos = "";
1060 
1061 		if (p->value)
1062 			pos = p->value;
1063 
1064 		json_object_add_value_string(job_opt_object, p->name, pos);
1065 	} else if (client->opt_lists) {
1066 		struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
1067 
1068 		flist_add_tail(&p->list, opt_list);
1069 	}
1070 }
1071 
handle_text(struct fio_client * client,struct fio_net_cmd * cmd)1072 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
1073 {
1074 	struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1075 	const char *buf = (const char *) pdu->buf;
1076 	const char *name;
1077 	int fio_unused ret;
1078 
1079 	name = client->name ? client->name : client->hostname;
1080 
1081 	if (!client->skip_newline)
1082 		fprintf(f_out, "<%s> ", name);
1083 	ret = fwrite(buf, pdu->buf_len, 1, f_out);
1084 	fflush(f_out);
1085 	client->skip_newline = strchr(buf, '\n') == NULL;
1086 }
1087 
convert_agg(struct disk_util_agg * agg)1088 static void convert_agg(struct disk_util_agg *agg)
1089 {
1090 	int i;
1091 
1092 	for (i = 0; i < 2; i++) {
1093 		agg->ios[i]	= le64_to_cpu(agg->ios[i]);
1094 		agg->merges[i]	= le64_to_cpu(agg->merges[i]);
1095 		agg->sectors[i]	= le64_to_cpu(agg->sectors[i]);
1096 		agg->ticks[i]	= le64_to_cpu(agg->ticks[i]);
1097 	}
1098 
1099 	agg->io_ticks		= le64_to_cpu(agg->io_ticks);
1100 	agg->time_in_queue	= le64_to_cpu(agg->time_in_queue);
1101 	agg->slavecount		= le32_to_cpu(agg->slavecount);
1102 	agg->max_util.u.f	= fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
1103 }
1104 
convert_dus(struct disk_util_stat * dus)1105 static void convert_dus(struct disk_util_stat *dus)
1106 {
1107 	int i;
1108 
1109 	for (i = 0; i < 2; i++) {
1110 		dus->s.ios[i]		= le64_to_cpu(dus->s.ios[i]);
1111 		dus->s.merges[i]	= le64_to_cpu(dus->s.merges[i]);
1112 		dus->s.sectors[i]	= le64_to_cpu(dus->s.sectors[i]);
1113 		dus->s.ticks[i]		= le64_to_cpu(dus->s.ticks[i]);
1114 	}
1115 
1116 	dus->s.io_ticks		= le64_to_cpu(dus->s.io_ticks);
1117 	dus->s.time_in_queue	= le64_to_cpu(dus->s.time_in_queue);
1118 	dus->s.msec		= le64_to_cpu(dus->s.msec);
1119 }
1120 
handle_du(struct fio_client * client,struct fio_net_cmd * cmd)1121 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
1122 {
1123 	struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1124 
1125 	if (!client->disk_stats_shown) {
1126 		client->disk_stats_shown = 1;
1127 		log_info("\nDisk stats (read/write):\n");
1128 	}
1129 
1130 	if (output_format & FIO_OUTPUT_JSON) {
1131 		struct json_object *duobj;
1132 		json_array_add_disk_util(&du->dus, &du->agg, du_array);
1133 		duobj = json_array_last_value_object(du_array);
1134 		json_object_add_client_info(duobj, client);
1135 	}
1136 	if (output_format & FIO_OUTPUT_TERSE)
1137 		print_disk_util(&du->dus, &du->agg, 1, NULL);
1138 	if (output_format & FIO_OUTPUT_NORMAL)
1139 		print_disk_util(&du->dus, &du->agg, 0, NULL);
1140 }
1141 
convert_jobs_eta(struct jobs_eta * je)1142 static void convert_jobs_eta(struct jobs_eta *je)
1143 {
1144 	int i;
1145 
1146 	je->nr_running		= le32_to_cpu(je->nr_running);
1147 	je->nr_ramp		= le32_to_cpu(je->nr_ramp);
1148 	je->nr_pending		= le32_to_cpu(je->nr_pending);
1149 	je->nr_setting_up	= le32_to_cpu(je->nr_setting_up);
1150 	je->files_open		= le32_to_cpu(je->files_open);
1151 
1152 	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1153 		je->m_rate[i]	= le64_to_cpu(je->m_rate[i]);
1154 		je->t_rate[i]	= le64_to_cpu(je->t_rate[i]);
1155 		je->m_iops[i]	= le32_to_cpu(je->m_iops[i]);
1156 		je->t_iops[i]	= le32_to_cpu(je->t_iops[i]);
1157 		je->rate[i]	= le64_to_cpu(je->rate[i]);
1158 		je->iops[i]	= le32_to_cpu(je->iops[i]);
1159 	}
1160 
1161 	je->elapsed_sec		= le64_to_cpu(je->elapsed_sec);
1162 	je->eta_sec		= le64_to_cpu(je->eta_sec);
1163 	je->nr_threads		= le32_to_cpu(je->nr_threads);
1164 	je->is_pow2		= le32_to_cpu(je->is_pow2);
1165 	je->unit_base		= le32_to_cpu(je->unit_base);
1166 }
1167 
fio_client_sum_jobs_eta(struct jobs_eta * dst,struct jobs_eta * je)1168 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
1169 {
1170 	int i;
1171 
1172 	dst->nr_running		+= je->nr_running;
1173 	dst->nr_ramp		+= je->nr_ramp;
1174 	dst->nr_pending		+= je->nr_pending;
1175 	dst->nr_setting_up	+= je->nr_setting_up;
1176 	dst->files_open		+= je->files_open;
1177 
1178 	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1179 		dst->m_rate[i]	+= je->m_rate[i];
1180 		dst->t_rate[i]	+= je->t_rate[i];
1181 		dst->m_iops[i]	+= je->m_iops[i];
1182 		dst->t_iops[i]	+= je->t_iops[i];
1183 		dst->rate[i]	+= je->rate[i];
1184 		dst->iops[i]	+= je->iops[i];
1185 	}
1186 
1187 	dst->elapsed_sec	+= je->elapsed_sec;
1188 
1189 	if (je->eta_sec > dst->eta_sec)
1190 		dst->eta_sec = je->eta_sec;
1191 
1192 	dst->nr_threads		+= je->nr_threads;
1193 
1194 	/*
1195 	 * This wont be correct for multiple strings, but at least it
1196 	 * works for the basic cases.
1197 	 */
1198 	strcpy((char *) dst->run_str, (char *) je->run_str);
1199 }
1200 
remove_reply_cmd(struct fio_client * client,struct fio_net_cmd * cmd)1201 static bool remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
1202 {
1203 	struct fio_net_cmd_reply *reply = NULL;
1204 	struct flist_head *entry;
1205 
1206 	flist_for_each(entry, &client->cmd_list) {
1207 		reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1208 
1209 		if (cmd->tag == (uintptr_t) reply)
1210 			break;
1211 
1212 		reply = NULL;
1213 	}
1214 
1215 	if (!reply) {
1216 		log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag);
1217 		return false;
1218 	}
1219 
1220 	flist_del(&reply->list);
1221 	cmd->tag = reply->saved_tag;
1222 	free(reply);
1223 	return true;
1224 }
1225 
fio_client_wait_for_reply(struct fio_client * client,uint64_t tag)1226 int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
1227 {
1228 	do {
1229 		struct fio_net_cmd_reply *reply = NULL;
1230 		struct flist_head *entry;
1231 
1232 		flist_for_each(entry, &client->cmd_list) {
1233 			reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1234 
1235 			if (tag == (uintptr_t) reply)
1236 				break;
1237 
1238 			reply = NULL;
1239 		}
1240 
1241 		if (!reply)
1242 			break;
1243 
1244 		usleep(1000);
1245 	} while (1);
1246 
1247 	return 0;
1248 }
1249 
handle_eta(struct fio_client * client,struct fio_net_cmd * cmd)1250 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
1251 {
1252 	struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1253 	struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
1254 
1255 	dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
1256 
1257 	assert(client->eta_in_flight == eta);
1258 
1259 	client->eta_in_flight = NULL;
1260 	flist_del_init(&client->eta_list);
1261 	client->eta_timeouts = 0;
1262 
1263 	if (client->ops->jobs_eta)
1264 		client->ops->jobs_eta(client, je);
1265 
1266 	fio_client_sum_jobs_eta(&eta->eta, je);
1267 	fio_client_dec_jobs_eta(eta, client->ops->eta);
1268 }
1269 
client_flush_hist_samples(FILE * f,int hist_coarseness,void * samples,uint64_t sample_size)1270 static void client_flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
1271 				      uint64_t sample_size)
1272 {
1273 	struct io_sample *s;
1274 	int log_offset;
1275 	uint64_t i, j, nr_samples;
1276 	struct io_u_plat_entry *entry;
1277 	unsigned int *io_u_plat;
1278 
1279 	int stride = 1 << hist_coarseness;
1280 
1281 	if (!sample_size)
1282 		return;
1283 
1284 	s = __get_sample(samples, 0, 0);
1285 	log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
1286 
1287 	nr_samples = sample_size / __log_entry_sz(log_offset);
1288 
1289 	for (i = 0; i < nr_samples; i++) {
1290 
1291 		s = (struct io_sample *)((char *)__get_sample(samples, log_offset, i) +
1292 			i * sizeof(struct io_u_plat_entry));
1293 
1294 		entry = s->data.plat_entry;
1295 		io_u_plat = entry->io_u_plat;
1296 
1297 		fprintf(f, "%lu, %u, %u, ", (unsigned long) s->time,
1298 						io_sample_ddir(s), s->bs);
1299 		for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) {
1300 			fprintf(f, "%lu, ", hist_sum(j, stride, io_u_plat, NULL));
1301 		}
1302 		fprintf(f, "%lu\n", (unsigned long)
1303 			hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat, NULL));
1304 
1305 	}
1306 }
1307 
fio_client_handle_iolog(struct fio_client * client,struct fio_net_cmd * cmd)1308 static int fio_client_handle_iolog(struct fio_client *client,
1309 				   struct fio_net_cmd *cmd)
1310 {
1311 	struct cmd_iolog_pdu *pdu;
1312 	bool store_direct;
1313 	char *log_pathname;
1314 
1315 	pdu = convert_iolog(cmd, &store_direct);
1316 	if (!pdu) {
1317 		log_err("fio: failed converting IO log\n");
1318 		return 1;
1319 	}
1320 
1321         /* allocate buffer big enough for next sprintf() call */
1322 	log_pathname = malloc(10 + strlen((char *)pdu->name) +
1323 			strlen(client->hostname));
1324 	if (!log_pathname) {
1325 		log_err("fio: memory allocation of unique pathname failed\n");
1326 		return -1;
1327 	}
1328 	/* generate a unique pathname for the log file using hostname */
1329 	sprintf(log_pathname, "%s.%s", pdu->name, client->hostname);
1330 
1331 	if (store_direct) {
1332 		ssize_t ret;
1333 		size_t sz;
1334 		int fd;
1335 
1336 		fd = open((const char *) log_pathname,
1337 				O_WRONLY | O_CREAT | O_TRUNC, 0644);
1338 		if (fd < 0) {
1339 			log_err("fio: open log %s: %s\n",
1340 				log_pathname, strerror(errno));
1341 			return 1;
1342 		}
1343 
1344 		sz = cmd->pdu_len - sizeof(*pdu);
1345 		ret = write(fd, pdu->samples, sz);
1346 		close(fd);
1347 
1348 		if (ret != sz) {
1349 			log_err("fio: short write on compressed log\n");
1350 			return 1;
1351 		}
1352 
1353 		return 0;
1354 	} else {
1355 		FILE *f;
1356 		f = fopen((const char *) log_pathname, "w");
1357 		if (!f) {
1358 			log_err("fio: fopen log %s : %s\n",
1359 				log_pathname, strerror(errno));
1360 			return 1;
1361 		}
1362 
1363 		if (pdu->log_type == IO_LOG_TYPE_HIST) {
1364 			client_flush_hist_samples(f, pdu->log_hist_coarseness, pdu->samples,
1365 					   pdu->nr_samples * sizeof(struct io_sample));
1366 		} else {
1367 			flush_samples(f, pdu->samples,
1368 					pdu->nr_samples * sizeof(struct io_sample));
1369 		}
1370 		fclose(f);
1371 		return 0;
1372 	}
1373 }
1374 
handle_probe(struct fio_client * client,struct fio_net_cmd * cmd)1375 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
1376 {
1377 	struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
1378 	const char *os, *arch;
1379 	char bit[16];
1380 
1381 	os = fio_get_os_string(probe->os);
1382 	if (!os)
1383 		os = "unknown";
1384 
1385 	arch = fio_get_arch_string(probe->arch);
1386 	if (!arch)
1387 		os = "unknown";
1388 
1389 	sprintf(bit, "%d-bit", probe->bpp * 8);
1390 	probe->flags = le64_to_cpu(probe->flags);
1391 
1392 	log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n",
1393 		probe->hostname, probe->bigendian, bit, os, arch,
1394 		probe->fio_version, (unsigned long) probe->flags);
1395 
1396 	if (!client->name)
1397 		client->name = strdup((char *) probe->hostname);
1398 }
1399 
handle_start(struct fio_client * client,struct fio_net_cmd * cmd)1400 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
1401 {
1402 	struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1403 
1404 	client->state = Client_started;
1405 	client->jobs = le32_to_cpu(pdu->jobs);
1406 	client->nr_stat = le32_to_cpu(pdu->stat_outputs);
1407 
1408 	if (client->jobs) {
1409 		int i;
1410 
1411 		if (client->opt_lists)
1412 			free(client->opt_lists);
1413 
1414 		client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
1415 		for (i = 0; i < client->jobs; i++)
1416 			INIT_FLIST_HEAD(&client->opt_lists[i]);
1417 	}
1418 
1419 	sum_stat_clients += client->nr_stat;
1420 }
1421 
handle_stop(struct fio_client * client,struct fio_net_cmd * cmd)1422 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
1423 {
1424 	if (client->error)
1425 		log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
1426 }
1427 
convert_stop(struct fio_net_cmd * cmd)1428 static void convert_stop(struct fio_net_cmd *cmd)
1429 {
1430 	struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1431 
1432 	pdu->error = le32_to_cpu(pdu->error);
1433 }
1434 
convert_text(struct fio_net_cmd * cmd)1435 static void convert_text(struct fio_net_cmd *cmd)
1436 {
1437 	struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1438 
1439 	pdu->level	= le32_to_cpu(pdu->level);
1440 	pdu->buf_len	= le32_to_cpu(pdu->buf_len);
1441 	pdu->log_sec	= le64_to_cpu(pdu->log_sec);
1442 	pdu->log_usec	= le64_to_cpu(pdu->log_usec);
1443 }
1444 
convert_iolog_gz(struct fio_net_cmd * cmd,struct cmd_iolog_pdu * pdu)1445 static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
1446 					      struct cmd_iolog_pdu *pdu)
1447 {
1448 #ifdef CONFIG_ZLIB
1449 	struct cmd_iolog_pdu *ret;
1450 	z_stream stream;
1451 	uint32_t nr_samples;
1452 	size_t total;
1453 	void *p;
1454 
1455 	stream.zalloc = Z_NULL;
1456 	stream.zfree = Z_NULL;
1457 	stream.opaque = Z_NULL;
1458 	stream.avail_in = 0;
1459 	stream.next_in = Z_NULL;
1460 
1461 	if (inflateInit(&stream) != Z_OK)
1462 		return NULL;
1463 
1464 	/*
1465 	 * Get header first, it's not compressed
1466 	 */
1467 	nr_samples = le64_to_cpu(pdu->nr_samples);
1468 
1469 	if (pdu->log_type == IO_LOG_TYPE_HIST)
1470 		total = nr_samples * (__log_entry_sz(le32_to_cpu(pdu->log_offset)) +
1471 					sizeof(struct io_u_plat_entry));
1472 	else
1473 		total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
1474 	ret = malloc(total + sizeof(*pdu));
1475 	ret->nr_samples = nr_samples;
1476 
1477 	memcpy(ret, pdu, sizeof(*pdu));
1478 
1479 	p = (void *) ret + sizeof(*pdu);
1480 
1481 	stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1482 	stream.next_in = (void *) pdu + sizeof(*pdu);
1483 	while (stream.avail_in) {
1484 		unsigned int this_chunk = 65536;
1485 		unsigned int this_len;
1486 		int err;
1487 
1488 		if (this_chunk > total)
1489 			this_chunk = total;
1490 
1491 		stream.avail_out = this_chunk;
1492 		stream.next_out = p;
1493 		err = inflate(&stream, Z_NO_FLUSH);
1494 		/* may be Z_OK, or Z_STREAM_END */
1495 		if (err < 0) {
1496 			log_err("fio: inflate error %d\n", err);
1497 			free(ret);
1498 			ret = NULL;
1499 			goto err;
1500 		}
1501 
1502 		this_len = this_chunk - stream.avail_out;
1503 		p += this_len;
1504 		total -= this_len;
1505 	}
1506 
1507 err:
1508 	inflateEnd(&stream);
1509 	return ret;
1510 #else
1511 	return NULL;
1512 #endif
1513 }
1514 
1515 /*
1516  * This has been compressed on the server side, since it can be big.
1517  * Uncompress here.
1518  */
convert_iolog(struct fio_net_cmd * cmd,bool * store_direct)1519 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd,
1520 					   bool *store_direct)
1521 {
1522 	struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1523 	struct cmd_iolog_pdu *ret;
1524 	uint64_t i;
1525 	int compressed;
1526 	void *samples;
1527 
1528 	*store_direct = false;
1529 
1530 	/*
1531 	 * Convert if compressed and we support it. If it's not
1532 	 * compressed, we need not do anything.
1533 	 */
1534 	compressed = le32_to_cpu(pdu->compressed);
1535 	if (compressed == XMIT_COMPRESSED) {
1536 #ifndef CONFIG_ZLIB
1537 		log_err("fio: server sent compressed data by mistake\n");
1538 		return NULL;
1539 #endif
1540 		ret = convert_iolog_gz(cmd, pdu);
1541 		if (!ret) {
1542 			log_err("fio: failed decompressing log\n");
1543 			return NULL;
1544 		}
1545 	} else if (compressed == STORE_COMPRESSED) {
1546 		*store_direct = true;
1547 		ret = pdu;
1548 	} else
1549 		ret = pdu;
1550 
1551 	ret->nr_samples		= le64_to_cpu(ret->nr_samples);
1552 	ret->thread_number	= le32_to_cpu(ret->thread_number);
1553 	ret->log_type		= le32_to_cpu(ret->log_type);
1554 	ret->compressed		= le32_to_cpu(ret->compressed);
1555 	ret->log_offset		= le32_to_cpu(ret->log_offset);
1556 	ret->log_hist_coarseness = le32_to_cpu(ret->log_hist_coarseness);
1557 
1558 	if (*store_direct)
1559 		return ret;
1560 
1561 	samples = &ret->samples[0];
1562 	for (i = 0; i < ret->nr_samples; i++) {
1563 		struct io_sample *s;
1564 
1565 		s = __get_sample(samples, ret->log_offset, i);
1566 		if (ret->log_type == IO_LOG_TYPE_HIST)
1567 			s = (struct io_sample *)((void *)s + sizeof(struct io_u_plat_entry) * i);
1568 
1569 		s->time		= le64_to_cpu(s->time);
1570 		s->data.val	= le64_to_cpu(s->data.val);
1571 		s->__ddir	= le32_to_cpu(s->__ddir);
1572 		s->bs		= le32_to_cpu(s->bs);
1573 
1574 		if (ret->log_offset) {
1575 			struct io_sample_offset *so = (void *) s;
1576 
1577 			so->offset = le64_to_cpu(so->offset);
1578 		}
1579 
1580 		if (ret->log_type == IO_LOG_TYPE_HIST) {
1581 			s->data.plat_entry = (struct io_u_plat_entry *)(((void *)s) + sizeof(*s));
1582 			s->data.plat_entry->list.next = NULL;
1583 			s->data.plat_entry->list.prev = NULL;
1584 		}
1585 	}
1586 
1587 	return ret;
1588 }
1589 
sendfile_reply(int fd,struct cmd_sendfile_reply * rep,size_t size,uint64_t tag)1590 static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep,
1591 			   size_t size, uint64_t tag)
1592 {
1593 	rep->error = cpu_to_le32(rep->error);
1594 	fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL);
1595 }
1596 
fio_send_file(struct fio_client * client,struct cmd_sendfile * pdu,uint64_t tag)1597 static int fio_send_file(struct fio_client *client, struct cmd_sendfile *pdu,
1598 			 uint64_t tag)
1599 {
1600 	struct cmd_sendfile_reply *rep;
1601 	struct stat sb;
1602 	size_t size;
1603 	int fd;
1604 
1605 	size = sizeof(*rep);
1606 	rep = malloc(size);
1607 
1608 	if (stat((char *)pdu->path, &sb) < 0) {
1609 fail:
1610 		rep->error = errno;
1611 		sendfile_reply(client->fd, rep, size, tag);
1612 		free(rep);
1613 		return 1;
1614 	}
1615 
1616 	size += sb.st_size;
1617 	rep = realloc(rep, size);
1618 	rep->size = cpu_to_le32((uint32_t) sb.st_size);
1619 
1620 	fd = open((char *)pdu->path, O_RDONLY);
1621 	if (fd == -1 )
1622 		goto fail;
1623 
1624 	rep->error = read_data(fd, &rep->data, sb.st_size);
1625 	sendfile_reply(client->fd, rep, size, tag);
1626 	free(rep);
1627 	close(fd);
1628 	return 0;
1629 }
1630 
fio_handle_client(struct fio_client * client)1631 int fio_handle_client(struct fio_client *client)
1632 {
1633 	struct client_ops *ops = client->ops;
1634 	struct fio_net_cmd *cmd;
1635 	int size;
1636 
1637 	dprint(FD_NET, "client: handle %s\n", client->hostname);
1638 
1639 	cmd = fio_net_recv_cmd(client->fd, true);
1640 	if (!cmd)
1641 		return 0;
1642 
1643 	dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1644 		fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1645 
1646 	switch (cmd->opcode) {
1647 	case FIO_NET_CMD_QUIT:
1648 		if (ops->quit)
1649 			ops->quit(client, cmd);
1650 		remove_client(client);
1651 		break;
1652 	case FIO_NET_CMD_TEXT:
1653 		convert_text(cmd);
1654 		ops->text(client, cmd);
1655 		break;
1656 	case FIO_NET_CMD_DU: {
1657 		struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1658 
1659 		convert_dus(&du->dus);
1660 		convert_agg(&du->agg);
1661 
1662 		ops->disk_util(client, cmd);
1663 		break;
1664 		}
1665 	case FIO_NET_CMD_TS: {
1666 		struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1667 
1668 		dprint(FD_NET, "client: ts->ss_state = %u\n", (unsigned int) le32_to_cpu(p->ts.ss_state));
1669 		if (le32_to_cpu(p->ts.ss_state) & __FIO_SS_DATA) {
1670 			dprint(FD_NET, "client: received steadystate ring buffers\n");
1671 
1672 			size = le64_to_cpu(p->ts.ss_dur);
1673 			p->ts.ss_iops_data = (uint64_t *) ((struct cmd_ts_pdu *)cmd->payload + 1);
1674 			p->ts.ss_bw_data = p->ts.ss_iops_data + size;
1675 		}
1676 
1677 		convert_ts(&p->ts, &p->ts);
1678 		convert_gs(&p->rs, &p->rs);
1679 
1680 		ops->thread_status(client, cmd);
1681 		break;
1682 		}
1683 	case FIO_NET_CMD_GS: {
1684 		struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1685 
1686 		convert_gs(gs, gs);
1687 
1688 		ops->group_stats(client, cmd);
1689 		break;
1690 		}
1691 	case FIO_NET_CMD_ETA: {
1692 		struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1693 
1694 		if (!remove_reply_cmd(client, cmd))
1695 			break;
1696 		convert_jobs_eta(je);
1697 		handle_eta(client, cmd);
1698 		break;
1699 		}
1700 	case FIO_NET_CMD_PROBE:
1701 		remove_reply_cmd(client, cmd);
1702 		ops->probe(client, cmd);
1703 		break;
1704 	case FIO_NET_CMD_SERVER_START:
1705 		client->state = Client_running;
1706 		if (ops->job_start)
1707 			ops->job_start(client, cmd);
1708 		break;
1709 	case FIO_NET_CMD_START: {
1710 		struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1711 
1712 		pdu->jobs = le32_to_cpu(pdu->jobs);
1713 		ops->start(client, cmd);
1714 		break;
1715 		}
1716 	case FIO_NET_CMD_STOP: {
1717 		struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1718 
1719 		convert_stop(cmd);
1720 		client->state = Client_stopped;
1721 		client->error = le32_to_cpu(pdu->error);
1722 		client->signal = le32_to_cpu(pdu->signal);
1723 		ops->stop(client, cmd);
1724 		break;
1725 		}
1726 	case FIO_NET_CMD_ADD_JOB: {
1727 		struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1728 
1729 		client->thread_number = le32_to_cpu(pdu->thread_number);
1730 		client->groupid = le32_to_cpu(pdu->groupid);
1731 
1732 		if (ops->add_job)
1733 			ops->add_job(client, cmd);
1734 		break;
1735 		}
1736 	case FIO_NET_CMD_IOLOG:
1737 		fio_client_handle_iolog(client, cmd);
1738 		break;
1739 	case FIO_NET_CMD_UPDATE_JOB:
1740 		ops->update_job(client, cmd);
1741 		remove_reply_cmd(client, cmd);
1742 		break;
1743 	case FIO_NET_CMD_VTRIGGER: {
1744 		struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
1745 		char buf[128];
1746 		int off = 0;
1747 
1748 		if (aux_path) {
1749 			strcpy(buf, aux_path);
1750 			off = strlen(buf);
1751 		}
1752 
1753 		__verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
1754 		exec_trigger(trigger_cmd);
1755 		break;
1756 		}
1757 	case FIO_NET_CMD_SENDFILE: {
1758 		struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload;
1759 		fio_send_file(client, pdu, cmd->tag);
1760 		break;
1761 		}
1762 	case FIO_NET_CMD_JOB_OPT: {
1763 		handle_job_opt(client, cmd);
1764 		break;
1765 	}
1766 	default:
1767 		log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1768 		break;
1769 	}
1770 
1771 	free(cmd);
1772 	return 1;
1773 }
1774 
fio_clients_send_trigger(const char * cmd)1775 int fio_clients_send_trigger(const char *cmd)
1776 {
1777 	struct flist_head *entry;
1778 	struct fio_client *client;
1779 	size_t slen;
1780 
1781 	dprint(FD_NET, "client: send vtrigger: %s\n", cmd);
1782 
1783 	if (!cmd)
1784 		slen = 0;
1785 	else
1786 		slen = strlen(cmd);
1787 
1788 	flist_for_each(entry, &client_list) {
1789 		struct cmd_vtrigger_pdu *pdu;
1790 
1791 		client = flist_entry(entry, struct fio_client, list);
1792 
1793 		pdu = malloc(sizeof(*pdu) + slen);
1794 		pdu->len = cpu_to_le16((uint16_t) slen);
1795 		if (slen)
1796 			memcpy(pdu->cmd, cmd, slen);
1797 		fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu,
1798 					sizeof(*pdu) + slen, NULL, NULL);
1799 		free(pdu);
1800 	}
1801 
1802 	return 0;
1803 }
1804 
request_client_etas(struct client_ops * ops)1805 static void request_client_etas(struct client_ops *ops)
1806 {
1807 	struct fio_client *client;
1808 	struct flist_head *entry;
1809 	struct client_eta *eta;
1810 	int skipped = 0;
1811 
1812 	dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1813 
1814 	eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
1815 	eta->pending = nr_clients;
1816 
1817 	flist_for_each(entry, &client_list) {
1818 		client = flist_entry(entry, struct fio_client, list);
1819 
1820 		if (!flist_empty(&client->eta_list)) {
1821 			skipped++;
1822 			continue;
1823 		}
1824 		if (client->state != Client_running)
1825 			continue;
1826 
1827 		assert(!client->eta_in_flight);
1828 		flist_add_tail(&client->eta_list, &eta_list);
1829 		client->eta_in_flight = eta;
1830 		fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1831 					(uintptr_t) eta, &client->cmd_list);
1832 	}
1833 
1834 	while (skipped--) {
1835 		if (!fio_client_dec_jobs_eta(eta, ops->eta))
1836 			break;
1837 	}
1838 
1839 	dprint(FD_NET, "client: requested eta tag %p\n", eta);
1840 }
1841 
1842 /*
1843  * A single SEND_ETA timeout isn't fatal. Attempt to recover.
1844  */
handle_cmd_timeout(struct fio_client * client,struct fio_net_cmd_reply * reply)1845 static int handle_cmd_timeout(struct fio_client *client,
1846 			      struct fio_net_cmd_reply *reply)
1847 {
1848 	flist_del(&reply->list);
1849 	free(reply);
1850 
1851 	if (reply->opcode != FIO_NET_CMD_SEND_ETA)
1852 		return 1;
1853 
1854 	log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
1855 
1856 	flist_del_init(&client->eta_list);
1857 	if (client->eta_in_flight) {
1858 		fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
1859 		client->eta_in_flight = NULL;
1860 	}
1861 
1862 	/*
1863 	 * If we fail 5 in a row, give up...
1864 	 */
1865 	if (client->eta_timeouts++ > 5)
1866 		return 1;
1867 
1868 	return 0;
1869 }
1870 
client_check_cmd_timeout(struct fio_client * client,struct timeval * now)1871 static int client_check_cmd_timeout(struct fio_client *client,
1872 				    struct timeval *now)
1873 {
1874 	struct fio_net_cmd_reply *reply;
1875 	struct flist_head *entry, *tmp;
1876 	int ret = 0;
1877 
1878 	flist_for_each_safe(entry, tmp, &client->cmd_list) {
1879 		reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1880 
1881 		if (mtime_since(&reply->tv, now) < FIO_NET_CLIENT_TIMEOUT)
1882 			continue;
1883 
1884 		if (!handle_cmd_timeout(client, reply))
1885 			continue;
1886 
1887 		log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1888 						fio_server_op(reply->opcode));
1889 		ret = 1;
1890 	}
1891 
1892 	return flist_empty(&client->cmd_list) && ret;
1893 }
1894 
fio_check_clients_timed_out(void)1895 static int fio_check_clients_timed_out(void)
1896 {
1897 	struct fio_client *client;
1898 	struct flist_head *entry, *tmp;
1899 	struct timeval tv;
1900 	int ret = 0;
1901 
1902 	fio_gettime(&tv, NULL);
1903 
1904 	flist_for_each_safe(entry, tmp, &client_list) {
1905 		client = flist_entry(entry, struct fio_client, list);
1906 
1907 		if (flist_empty(&client->cmd_list))
1908 			continue;
1909 
1910 		if (!client_check_cmd_timeout(client, &tv))
1911 			continue;
1912 
1913 		if (client->ops->timed_out)
1914 			client->ops->timed_out(client);
1915 		else
1916 			log_err("fio: client %s timed out\n", client->hostname);
1917 
1918 		client->error = ETIMEDOUT;
1919 		remove_client(client);
1920 		ret = 1;
1921 	}
1922 
1923 	return ret;
1924 }
1925 
fio_handle_clients(struct client_ops * ops)1926 int fio_handle_clients(struct client_ops *ops)
1927 {
1928 	struct pollfd *pfds;
1929 	int i, ret = 0, retval = 0;
1930 
1931 	fio_gettime(&eta_tv, NULL);
1932 
1933 	pfds = malloc(nr_clients * sizeof(struct pollfd));
1934 
1935 	init_thread_stat(&client_ts);
1936 	init_group_run_stat(&client_gs);
1937 
1938 	while (!exit_backend && nr_clients) {
1939 		struct flist_head *entry, *tmp;
1940 		struct fio_client *client;
1941 
1942 		i = 0;
1943 		flist_for_each_safe(entry, tmp, &client_list) {
1944 			client = flist_entry(entry, struct fio_client, list);
1945 
1946 			if (!client->sent_job && !client->ops->stay_connected &&
1947 			    flist_empty(&client->cmd_list)) {
1948 				remove_client(client);
1949 				continue;
1950 			}
1951 
1952 			pfds[i].fd = client->fd;
1953 			pfds[i].events = POLLIN;
1954 			i++;
1955 		}
1956 
1957 		if (!nr_clients)
1958 			break;
1959 
1960 		assert(i == nr_clients);
1961 
1962 		do {
1963 			struct timeval tv;
1964 			int timeout;
1965 
1966 			fio_gettime(&tv, NULL);
1967 			if (mtime_since(&eta_tv, &tv) >= 900) {
1968 				request_client_etas(ops);
1969 				memcpy(&eta_tv, &tv, sizeof(tv));
1970 
1971 				if (fio_check_clients_timed_out())
1972 					break;
1973 			}
1974 
1975 			check_trigger_file();
1976 
1977 			timeout = min(100u, ops->eta_msec);
1978 
1979 			ret = poll(pfds, nr_clients, timeout);
1980 			if (ret < 0) {
1981 				if (errno == EINTR)
1982 					continue;
1983 				log_err("fio: poll clients: %s\n", strerror(errno));
1984 				break;
1985 			} else if (!ret)
1986 				continue;
1987 		} while (ret <= 0);
1988 
1989 		for (i = 0; i < nr_clients; i++) {
1990 			if (!(pfds[i].revents & POLLIN))
1991 				continue;
1992 
1993 			client = find_client_by_fd(pfds[i].fd);
1994 			if (!client) {
1995 				log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd);
1996 				continue;
1997 			}
1998 			if (!fio_handle_client(client)) {
1999 				log_info("client: host=%s disconnected\n",
2000 						client->hostname);
2001 				remove_client(client);
2002 				retval = 1;
2003 			} else if (client->error)
2004 				retval = 1;
2005 			fio_put_client(client);
2006 		}
2007 	}
2008 
2009 	fio_client_json_fini();
2010 
2011 	free(pfds);
2012 	return retval || error_clients;
2013 }
2014