fio: add check rate flag
[fio.git] / client.c
... / ...
CommitLineData
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
28static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
29static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd);
30static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd);
31static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
32static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd);
33static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd);
34static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd);
35
36static void convert_text(struct fio_net_cmd *cmd);
37
38struct 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
51static struct timespec eta_ts;
52
53static FLIST_HEAD(client_list);
54static FLIST_HEAD(eta_list);
55
56static FLIST_HEAD(arg_list);
57
58struct thread_stat client_ts;
59struct group_run_stats client_gs;
60int sum_stat_clients;
61
62static int sum_stat_nr;
63static struct json_object *root = NULL;
64static struct json_object *job_opt_object = NULL;
65static struct json_array *clients_array = NULL;
66static struct json_array *du_array = NULL;
67
68static 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)
73static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
74
75static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *, bool *);
76
77static 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
85static 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
91static 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
99static 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
123static 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
148static 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
162static 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
180void 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
209static 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
220static 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
238static 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
266struct fio_client *fio_get_client(struct fio_client *client)
267{
268 client->refs++;
269 return client;
270}
271
272static 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
283void 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
303struct 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 = true;
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;
345err:
346 free(client);
347 return NULL;
348}
349
350int 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
375int 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
423static 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
438static 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
460static 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
501static 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
531int 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
555int fio_client_terminate(struct fio_client *client)
556{
557 return fio_net_send_quit(client->fd);
558}
559
560static 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
573static void sig_int(int sig)
574{
575 dprint(FD_NET, "client: got signal %d\n", sig);
576 fio_clients_terminate();
577}
578
579static 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
607static 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
655int 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
686int 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
692int 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
715static 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 = true;
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 */
741static 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 = true;
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
791int 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 = true;
803
804 return ret;
805}
806
807static 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
813int 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
847int 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
859static 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
872static 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 convert_io_stat(&dst->iops_stat[i], &src->iops_stat[i]);
889 }
890
891 dst->usr_time = le64_to_cpu(src->usr_time);
892 dst->sys_time = le64_to_cpu(src->sys_time);
893 dst->ctx = le64_to_cpu(src->ctx);
894 dst->minf = le64_to_cpu(src->minf);
895 dst->majf = le64_to_cpu(src->majf);
896 dst->clat_percentiles = le32_to_cpu(src->clat_percentiles);
897 dst->lat_percentiles = le32_to_cpu(src->lat_percentiles);
898 dst->percentile_precision = le64_to_cpu(src->percentile_precision);
899
900 for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
901 fio_fp64_t *fps = &src->percentile_list[i];
902 fio_fp64_t *fpd = &dst->percentile_list[i];
903
904 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
905 }
906
907 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
908 dst->io_u_map[i] = le32_to_cpu(src->io_u_map[i]);
909 dst->io_u_submit[i] = le32_to_cpu(src->io_u_submit[i]);
910 dst->io_u_complete[i] = le32_to_cpu(src->io_u_complete[i]);
911 }
912
913 for (i = 0; i < FIO_IO_U_LAT_N_NR; i++)
914 dst->io_u_lat_n[i] = le32_to_cpu(src->io_u_lat_n[i]);
915 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
916 dst->io_u_lat_u[i] = le32_to_cpu(src->io_u_lat_u[i]);
917 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
918 dst->io_u_lat_m[i] = le32_to_cpu(src->io_u_lat_m[i]);
919
920 for (i = 0; i < DDIR_RWDIR_CNT; i++)
921 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
922 dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
923
924 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
925 dst->total_io_u[i] = le64_to_cpu(src->total_io_u[i]);
926 dst->short_io_u[i] = le64_to_cpu(src->short_io_u[i]);
927 dst->drop_io_u[i] = le64_to_cpu(src->drop_io_u[i]);
928 }
929
930 dst->total_submit = le64_to_cpu(src->total_submit);
931 dst->total_complete = le64_to_cpu(src->total_complete);
932
933 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
934 dst->io_bytes[i] = le64_to_cpu(src->io_bytes[i]);
935 dst->runtime[i] = le64_to_cpu(src->runtime[i]);
936 }
937
938 dst->total_run_time = le64_to_cpu(src->total_run_time);
939 dst->continue_on_error = le16_to_cpu(src->continue_on_error);
940 dst->total_err_count = le64_to_cpu(src->total_err_count);
941 dst->first_error = le32_to_cpu(src->first_error);
942 dst->kb_base = le32_to_cpu(src->kb_base);
943 dst->unit_base = le32_to_cpu(src->unit_base);
944
945 dst->sig_figs = le32_to_cpu(src->sig_figs);
946
947 dst->latency_depth = le32_to_cpu(src->latency_depth);
948 dst->latency_target = le64_to_cpu(src->latency_target);
949 dst->latency_window = le64_to_cpu(src->latency_window);
950 dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
951
952 dst->nr_block_infos = le64_to_cpu(src->nr_block_infos);
953 for (i = 0; i < dst->nr_block_infos; i++)
954 dst->block_infos[i] = le32_to_cpu(src->block_infos[i]);
955
956 dst->ss_dur = le64_to_cpu(src->ss_dur);
957 dst->ss_state = le32_to_cpu(src->ss_state);
958 dst->ss_head = le32_to_cpu(src->ss_head);
959 dst->ss_limit.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_limit.u.i));
960 dst->ss_slope.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_slope.u.i));
961 dst->ss_deviation.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_deviation.u.i));
962 dst->ss_criterion.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_criterion.u.i));
963
964 if (dst->ss_state & __FIO_SS_DATA) {
965 for (i = 0; i < dst->ss_dur; i++ ) {
966 dst->ss_iops_data[i] = le64_to_cpu(src->ss_iops_data[i]);
967 dst->ss_bw_data[i] = le64_to_cpu(src->ss_bw_data[i]);
968 }
969 }
970}
971
972static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
973{
974 int i;
975
976 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
977 dst->max_run[i] = le64_to_cpu(src->max_run[i]);
978 dst->min_run[i] = le64_to_cpu(src->min_run[i]);
979 dst->max_bw[i] = le64_to_cpu(src->max_bw[i]);
980 dst->min_bw[i] = le64_to_cpu(src->min_bw[i]);
981 dst->iobytes[i] = le64_to_cpu(src->iobytes[i]);
982 dst->agg[i] = le64_to_cpu(src->agg[i]);
983 }
984
985 dst->kb_base = le32_to_cpu(src->kb_base);
986 dst->unit_base = le32_to_cpu(src->unit_base);
987 dst->sig_figs = le32_to_cpu(src->sig_figs);
988 dst->groupid = le32_to_cpu(src->groupid);
989 dst->unified_rw_rep = le32_to_cpu(src->unified_rw_rep);
990}
991
992static void json_object_add_client_info(struct json_object *obj,
993 struct fio_client *client)
994{
995 const char *hostname = client->hostname ? client->hostname : "";
996
997 json_object_add_value_string(obj, "hostname", hostname);
998 json_object_add_value_int(obj, "port", client->port);
999}
1000
1001static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
1002{
1003 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1004 struct flist_head *opt_list = NULL;
1005 struct json_object *tsobj;
1006
1007 if (client->opt_lists && p->ts.thread_number <= client->jobs)
1008 opt_list = &client->opt_lists[p->ts.thread_number - 1];
1009
1010 tsobj = show_thread_status(&p->ts, &p->rs, opt_list, NULL);
1011 client->did_stat = true;
1012 if (tsobj) {
1013 json_object_add_client_info(tsobj, client);
1014 json_array_add_value_object(clients_array, tsobj);
1015 }
1016
1017 if (sum_stat_clients <= 1)
1018 return;
1019
1020 sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1);
1021 sum_group_stats(&client_gs, &p->rs);
1022
1023 client_ts.members++;
1024 client_ts.thread_number = p->ts.thread_number;
1025 client_ts.groupid = p->ts.groupid;
1026 client_ts.unified_rw_rep = p->ts.unified_rw_rep;
1027
1028 if (++sum_stat_nr == sum_stat_clients) {
1029 strcpy(client_ts.name, "All clients");
1030 tsobj = show_thread_status(&client_ts, &client_gs, NULL, NULL);
1031 if (tsobj) {
1032 json_object_add_client_info(tsobj, client);
1033 json_array_add_value_object(clients_array, tsobj);
1034 }
1035 }
1036}
1037
1038static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
1039{
1040 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1041
1042 if (output_format & FIO_OUTPUT_NORMAL)
1043 show_group_stats(gs, NULL);
1044}
1045
1046static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
1047{
1048 struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
1049 struct print_option *p;
1050
1051 if (!job_opt_object)
1052 return;
1053
1054 pdu->global = le16_to_cpu(pdu->global);
1055 pdu->truncated = le16_to_cpu(pdu->truncated);
1056 pdu->groupid = le32_to_cpu(pdu->groupid);
1057
1058 p = malloc(sizeof(*p));
1059 p->name = strdup((char *) pdu->name);
1060 if (pdu->value[0] != '\0')
1061 p->value = strdup((char *) pdu->value);
1062 else
1063 p->value = NULL;
1064
1065 if (pdu->global) {
1066 const char *pos = "";
1067
1068 if (p->value)
1069 pos = p->value;
1070
1071 json_object_add_value_string(job_opt_object, p->name, pos);
1072 } else if (client->opt_lists) {
1073 struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
1074
1075 flist_add_tail(&p->list, opt_list);
1076 }
1077}
1078
1079static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
1080{
1081 struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1082 const char *buf = (const char *) pdu->buf;
1083 const char *name;
1084 int fio_unused ret;
1085
1086 name = client->name ? client->name : client->hostname;
1087
1088 if (!client->skip_newline)
1089 fprintf(f_out, "<%s> ", name);
1090 ret = fwrite(buf, pdu->buf_len, 1, f_out);
1091 fflush(f_out);
1092 client->skip_newline = strchr(buf, '\n') == NULL;
1093}
1094
1095static void convert_agg(struct disk_util_agg *agg)
1096{
1097 int i;
1098
1099 for (i = 0; i < 2; i++) {
1100 agg->ios[i] = le64_to_cpu(agg->ios[i]);
1101 agg->merges[i] = le64_to_cpu(agg->merges[i]);
1102 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
1103 agg->ticks[i] = le64_to_cpu(agg->ticks[i]);
1104 }
1105
1106 agg->io_ticks = le64_to_cpu(agg->io_ticks);
1107 agg->time_in_queue = le64_to_cpu(agg->time_in_queue);
1108 agg->slavecount = le32_to_cpu(agg->slavecount);
1109 agg->max_util.u.f = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
1110}
1111
1112static void convert_dus(struct disk_util_stat *dus)
1113{
1114 int i;
1115
1116 for (i = 0; i < 2; i++) {
1117 dus->s.ios[i] = le64_to_cpu(dus->s.ios[i]);
1118 dus->s.merges[i] = le64_to_cpu(dus->s.merges[i]);
1119 dus->s.sectors[i] = le64_to_cpu(dus->s.sectors[i]);
1120 dus->s.ticks[i] = le64_to_cpu(dus->s.ticks[i]);
1121 }
1122
1123 dus->s.io_ticks = le64_to_cpu(dus->s.io_ticks);
1124 dus->s.time_in_queue = le64_to_cpu(dus->s.time_in_queue);
1125 dus->s.msec = le64_to_cpu(dus->s.msec);
1126}
1127
1128static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
1129{
1130 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1131
1132 if (!client->disk_stats_shown) {
1133 client->disk_stats_shown = true;
1134 log_info("\nDisk stats (read/write):\n");
1135 }
1136
1137 if (output_format & FIO_OUTPUT_JSON) {
1138 struct json_object *duobj;
1139 json_array_add_disk_util(&du->dus, &du->agg, du_array);
1140 duobj = json_array_last_value_object(du_array);
1141 json_object_add_client_info(duobj, client);
1142 }
1143 if (output_format & FIO_OUTPUT_TERSE)
1144 print_disk_util(&du->dus, &du->agg, 1, NULL);
1145 if (output_format & FIO_OUTPUT_NORMAL)
1146 print_disk_util(&du->dus, &du->agg, 0, NULL);
1147}
1148
1149static void convert_jobs_eta(struct jobs_eta *je)
1150{
1151 int i;
1152
1153 je->nr_running = le32_to_cpu(je->nr_running);
1154 je->nr_ramp = le32_to_cpu(je->nr_ramp);
1155 je->nr_pending = le32_to_cpu(je->nr_pending);
1156 je->nr_setting_up = le32_to_cpu(je->nr_setting_up);
1157 je->files_open = le32_to_cpu(je->files_open);
1158
1159 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1160 je->m_rate[i] = le64_to_cpu(je->m_rate[i]);
1161 je->t_rate[i] = le64_to_cpu(je->t_rate[i]);
1162 je->m_iops[i] = le32_to_cpu(je->m_iops[i]);
1163 je->t_iops[i] = le32_to_cpu(je->t_iops[i]);
1164 je->rate[i] = le64_to_cpu(je->rate[i]);
1165 je->iops[i] = le32_to_cpu(je->iops[i]);
1166 }
1167
1168 je->elapsed_sec = le64_to_cpu(je->elapsed_sec);
1169 je->eta_sec = le64_to_cpu(je->eta_sec);
1170 je->nr_threads = le32_to_cpu(je->nr_threads);
1171 je->is_pow2 = le32_to_cpu(je->is_pow2);
1172 je->unit_base = le32_to_cpu(je->unit_base);
1173 je->sig_figs = le32_to_cpu(je->sig_figs);
1174}
1175
1176void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
1177{
1178 int i;
1179
1180 dst->nr_running += je->nr_running;
1181 dst->nr_ramp += je->nr_ramp;
1182 dst->nr_pending += je->nr_pending;
1183 dst->nr_setting_up += je->nr_setting_up;
1184 dst->files_open += je->files_open;
1185
1186 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1187 dst->m_rate[i] += je->m_rate[i];
1188 dst->t_rate[i] += je->t_rate[i];
1189 dst->m_iops[i] += je->m_iops[i];
1190 dst->t_iops[i] += je->t_iops[i];
1191 dst->rate[i] += je->rate[i];
1192 dst->iops[i] += je->iops[i];
1193 }
1194
1195 dst->elapsed_sec += je->elapsed_sec;
1196
1197 if (je->eta_sec > dst->eta_sec)
1198 dst->eta_sec = je->eta_sec;
1199
1200 dst->nr_threads += je->nr_threads;
1201
1202 /*
1203 * This wont be correct for multiple strings, but at least it
1204 * works for the basic cases.
1205 */
1206 strcpy((char *) dst->run_str, (char *) je->run_str);
1207}
1208
1209static bool remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
1210{
1211 struct fio_net_cmd_reply *reply = NULL;
1212 struct flist_head *entry;
1213
1214 flist_for_each(entry, &client->cmd_list) {
1215 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1216
1217 if (cmd->tag == (uintptr_t) reply)
1218 break;
1219
1220 reply = NULL;
1221 }
1222
1223 if (!reply) {
1224 log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag);
1225 return false;
1226 }
1227
1228 flist_del(&reply->list);
1229 cmd->tag = reply->saved_tag;
1230 free(reply);
1231 return true;
1232}
1233
1234int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
1235{
1236 do {
1237 struct fio_net_cmd_reply *reply = NULL;
1238 struct flist_head *entry;
1239
1240 flist_for_each(entry, &client->cmd_list) {
1241 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1242
1243 if (tag == (uintptr_t) reply)
1244 break;
1245
1246 reply = NULL;
1247 }
1248
1249 if (!reply)
1250 break;
1251
1252 usleep(1000);
1253 } while (1);
1254
1255 return 0;
1256}
1257
1258static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
1259{
1260 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1261 struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
1262
1263 dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
1264
1265 assert(client->eta_in_flight == eta);
1266
1267 client->eta_in_flight = NULL;
1268 flist_del_init(&client->eta_list);
1269 client->eta_timeouts = 0;
1270
1271 if (client->ops->jobs_eta)
1272 client->ops->jobs_eta(client, je);
1273
1274 fio_client_sum_jobs_eta(&eta->eta, je);
1275 fio_client_dec_jobs_eta(eta, client->ops->eta);
1276}
1277
1278static void client_flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
1279 uint64_t sample_size)
1280{
1281 struct io_sample *s;
1282 int log_offset;
1283 uint64_t i, j, nr_samples;
1284 struct io_u_plat_entry *entry;
1285 unsigned int *io_u_plat;
1286
1287 int stride = 1 << hist_coarseness;
1288
1289 if (!sample_size)
1290 return;
1291
1292 s = __get_sample(samples, 0, 0);
1293 log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
1294
1295 nr_samples = sample_size / __log_entry_sz(log_offset);
1296
1297 for (i = 0; i < nr_samples; i++) {
1298
1299 s = (struct io_sample *)((char *)__get_sample(samples, log_offset, i) +
1300 i * sizeof(struct io_u_plat_entry));
1301
1302 entry = s->data.plat_entry;
1303 io_u_plat = entry->io_u_plat;
1304
1305 fprintf(f, "%lu, %u, %u, ", (unsigned long) s->time,
1306 io_sample_ddir(s), s->bs);
1307 for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) {
1308 fprintf(f, "%lu, ", hist_sum(j, stride, io_u_plat, NULL));
1309 }
1310 fprintf(f, "%lu\n", (unsigned long)
1311 hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat, NULL));
1312
1313 }
1314}
1315
1316static int fio_client_handle_iolog(struct fio_client *client,
1317 struct fio_net_cmd *cmd)
1318{
1319 struct cmd_iolog_pdu *pdu = NULL;
1320 bool store_direct;
1321 char *log_pathname = NULL;
1322 int ret = 0;
1323
1324 pdu = convert_iolog(cmd, &store_direct);
1325 if (!pdu) {
1326 log_err("fio: failed converting IO log\n");
1327 ret = 1;
1328 goto out;
1329 }
1330
1331 /* allocate buffer big enough for next sprintf() call */
1332 log_pathname = malloc(10 + strlen((char *)pdu->name) +
1333 strlen(client->hostname));
1334 if (!log_pathname) {
1335 log_err("fio: memory allocation of unique pathname failed\n");
1336 ret = -1;
1337 goto out;
1338 }
1339 /* generate a unique pathname for the log file using hostname */
1340 sprintf(log_pathname, "%s.%s", pdu->name, client->hostname);
1341
1342 if (store_direct) {
1343 ssize_t ret;
1344 size_t sz;
1345 int fd;
1346
1347 fd = open((const char *) log_pathname,
1348 O_WRONLY | O_CREAT | O_TRUNC, 0644);
1349 if (fd < 0) {
1350 log_err("fio: open log %s: %s\n",
1351 log_pathname, strerror(errno));
1352 ret = 1;
1353 goto out;
1354 }
1355
1356 sz = cmd->pdu_len - sizeof(*pdu);
1357 ret = write(fd, pdu->samples, sz);
1358 close(fd);
1359
1360 if (ret != sz) {
1361 log_err("fio: short write on compressed log\n");
1362 ret = 1;
1363 goto out;
1364 }
1365
1366 ret = 0;
1367 } else {
1368 FILE *f;
1369 f = fopen((const char *) log_pathname, "w");
1370 if (!f) {
1371 log_err("fio: fopen log %s : %s\n",
1372 log_pathname, strerror(errno));
1373 ret = 1;
1374 goto out;
1375 }
1376
1377 if (pdu->log_type == IO_LOG_TYPE_HIST) {
1378 client_flush_hist_samples(f, pdu->log_hist_coarseness, pdu->samples,
1379 pdu->nr_samples * sizeof(struct io_sample));
1380 } else {
1381 flush_samples(f, pdu->samples,
1382 pdu->nr_samples * sizeof(struct io_sample));
1383 }
1384 fclose(f);
1385 ret = 0;
1386 }
1387
1388out:
1389 if (pdu && pdu != (void *) cmd->payload)
1390 free(pdu);
1391
1392 if (log_pathname)
1393 free(log_pathname);
1394
1395 return ret;
1396}
1397
1398static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
1399{
1400 struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
1401 const char *os, *arch;
1402 char bit[16];
1403
1404 os = fio_get_os_string(probe->os);
1405 if (!os)
1406 os = "unknown";
1407
1408 arch = fio_get_arch_string(probe->arch);
1409 if (!arch)
1410 os = "unknown";
1411
1412 sprintf(bit, "%d-bit", probe->bpp * 8);
1413 probe->flags = le64_to_cpu(probe->flags);
1414
1415 log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n",
1416 probe->hostname, probe->bigendian, bit, os, arch,
1417 probe->fio_version, (unsigned long) probe->flags);
1418
1419 if (!client->name)
1420 client->name = strdup((char *) probe->hostname);
1421}
1422
1423static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
1424{
1425 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1426
1427 client->state = Client_started;
1428 client->jobs = le32_to_cpu(pdu->jobs);
1429 client->nr_stat = le32_to_cpu(pdu->stat_outputs);
1430
1431 if (client->jobs) {
1432 int i;
1433
1434 if (client->opt_lists)
1435 free(client->opt_lists);
1436
1437 client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
1438 for (i = 0; i < client->jobs; i++)
1439 INIT_FLIST_HEAD(&client->opt_lists[i]);
1440 }
1441
1442 sum_stat_clients += client->nr_stat;
1443}
1444
1445static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
1446{
1447 if (client->error)
1448 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
1449}
1450
1451static void convert_stop(struct fio_net_cmd *cmd)
1452{
1453 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1454
1455 pdu->error = le32_to_cpu(pdu->error);
1456}
1457
1458static void convert_text(struct fio_net_cmd *cmd)
1459{
1460 struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1461
1462 pdu->level = le32_to_cpu(pdu->level);
1463 pdu->buf_len = le32_to_cpu(pdu->buf_len);
1464 pdu->log_sec = le64_to_cpu(pdu->log_sec);
1465 pdu->log_usec = le64_to_cpu(pdu->log_usec);
1466}
1467
1468static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
1469 struct cmd_iolog_pdu *pdu)
1470{
1471#ifdef CONFIG_ZLIB
1472 struct cmd_iolog_pdu *ret;
1473 z_stream stream;
1474 uint32_t nr_samples;
1475 size_t total;
1476 char *p;
1477
1478 stream.zalloc = Z_NULL;
1479 stream.zfree = Z_NULL;
1480 stream.opaque = Z_NULL;
1481 stream.avail_in = 0;
1482 stream.next_in = Z_NULL;
1483
1484 if (inflateInit(&stream) != Z_OK)
1485 return NULL;
1486
1487 /*
1488 * Get header first, it's not compressed
1489 */
1490 nr_samples = le64_to_cpu(pdu->nr_samples);
1491
1492 if (pdu->log_type == IO_LOG_TYPE_HIST)
1493 total = nr_samples * (__log_entry_sz(le32_to_cpu(pdu->log_offset)) +
1494 sizeof(struct io_u_plat_entry));
1495 else
1496 total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
1497 ret = malloc(total + sizeof(*pdu));
1498 ret->nr_samples = nr_samples;
1499
1500 memcpy(ret, pdu, sizeof(*pdu));
1501
1502 p = (char *) ret + sizeof(*pdu);
1503
1504 stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1505 stream.next_in = (void *)((char *) pdu + sizeof(*pdu));
1506 while (stream.avail_in) {
1507 unsigned int this_chunk = 65536;
1508 unsigned int this_len;
1509 int err;
1510
1511 if (this_chunk > total)
1512 this_chunk = total;
1513
1514 stream.avail_out = this_chunk;
1515 stream.next_out = (void *)p;
1516 err = inflate(&stream, Z_NO_FLUSH);
1517 /* may be Z_OK, or Z_STREAM_END */
1518 if (err < 0) {
1519 log_err("fio: inflate error %d\n", err);
1520 free(ret);
1521 ret = NULL;
1522 goto err;
1523 }
1524
1525 this_len = this_chunk - stream.avail_out;
1526 p += this_len;
1527 total -= this_len;
1528 }
1529
1530err:
1531 inflateEnd(&stream);
1532 return ret;
1533#else
1534 return NULL;
1535#endif
1536}
1537
1538/*
1539 * This has been compressed on the server side, since it can be big.
1540 * Uncompress here.
1541 */
1542static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd,
1543 bool *store_direct)
1544{
1545 struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1546 struct cmd_iolog_pdu *ret;
1547 uint64_t i;
1548 int compressed;
1549 void *samples;
1550
1551 *store_direct = false;
1552
1553 /*
1554 * Convert if compressed and we support it. If it's not
1555 * compressed, we need not do anything.
1556 */
1557 compressed = le32_to_cpu(pdu->compressed);
1558 if (compressed == XMIT_COMPRESSED) {
1559#ifndef CONFIG_ZLIB
1560 log_err("fio: server sent compressed data by mistake\n");
1561 return NULL;
1562#endif
1563 ret = convert_iolog_gz(cmd, pdu);
1564 if (!ret) {
1565 log_err("fio: failed decompressing log\n");
1566 return NULL;
1567 }
1568 } else if (compressed == STORE_COMPRESSED) {
1569 *store_direct = true;
1570 ret = pdu;
1571 } else
1572 ret = pdu;
1573
1574 ret->nr_samples = le64_to_cpu(ret->nr_samples);
1575 ret->thread_number = le32_to_cpu(ret->thread_number);
1576 ret->log_type = le32_to_cpu(ret->log_type);
1577 ret->compressed = le32_to_cpu(ret->compressed);
1578 ret->log_offset = le32_to_cpu(ret->log_offset);
1579 ret->log_hist_coarseness = le32_to_cpu(ret->log_hist_coarseness);
1580
1581 if (*store_direct)
1582 return ret;
1583
1584 samples = &ret->samples[0];
1585 for (i = 0; i < ret->nr_samples; i++) {
1586 struct io_sample *s;
1587
1588 s = __get_sample(samples, ret->log_offset, i);
1589 if (ret->log_type == IO_LOG_TYPE_HIST)
1590 s = (struct io_sample *)((char *)s + sizeof(struct io_u_plat_entry) * i);
1591
1592 s->time = le64_to_cpu(s->time);
1593 s->data.val = le64_to_cpu(s->data.val);
1594 s->__ddir = le32_to_cpu(s->__ddir);
1595 s->bs = le32_to_cpu(s->bs);
1596
1597 if (ret->log_offset) {
1598 struct io_sample_offset *so = (void *) s;
1599
1600 so->offset = le64_to_cpu(so->offset);
1601 }
1602
1603 if (ret->log_type == IO_LOG_TYPE_HIST) {
1604 s->data.plat_entry = (struct io_u_plat_entry *)(((char *)s) + sizeof(*s));
1605 s->data.plat_entry->list.next = NULL;
1606 s->data.plat_entry->list.prev = NULL;
1607 }
1608 }
1609
1610 return ret;
1611}
1612
1613static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep,
1614 size_t size, uint64_t tag)
1615{
1616 rep->error = cpu_to_le32(rep->error);
1617 fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL);
1618}
1619
1620static int fio_send_file(struct fio_client *client, struct cmd_sendfile *pdu,
1621 uint64_t tag)
1622{
1623 struct cmd_sendfile_reply *rep;
1624 struct stat sb;
1625 size_t size;
1626 int fd;
1627
1628 size = sizeof(*rep);
1629 rep = malloc(size);
1630
1631 if (stat((char *)pdu->path, &sb) < 0) {
1632fail:
1633 rep->error = errno;
1634 sendfile_reply(client->fd, rep, size, tag);
1635 free(rep);
1636 return 1;
1637 }
1638
1639 size += sb.st_size;
1640 rep = realloc(rep, size);
1641 rep->size = cpu_to_le32((uint32_t) sb.st_size);
1642
1643 fd = open((char *)pdu->path, O_RDONLY);
1644 if (fd == -1 )
1645 goto fail;
1646
1647 rep->error = read_data(fd, &rep->data, sb.st_size);
1648 sendfile_reply(client->fd, rep, size, tag);
1649 free(rep);
1650 close(fd);
1651 return 0;
1652}
1653
1654int fio_handle_client(struct fio_client *client)
1655{
1656 struct client_ops *ops = client->ops;
1657 struct fio_net_cmd *cmd;
1658 int size;
1659
1660 dprint(FD_NET, "client: handle %s\n", client->hostname);
1661
1662 cmd = fio_net_recv_cmd(client->fd, true);
1663 if (!cmd)
1664 return 0;
1665
1666 dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1667 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1668
1669 switch (cmd->opcode) {
1670 case FIO_NET_CMD_QUIT:
1671 if (ops->quit)
1672 ops->quit(client, cmd);
1673 remove_client(client);
1674 break;
1675 case FIO_NET_CMD_TEXT:
1676 convert_text(cmd);
1677 ops->text(client, cmd);
1678 break;
1679 case FIO_NET_CMD_DU: {
1680 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1681
1682 convert_dus(&du->dus);
1683 convert_agg(&du->agg);
1684
1685 ops->disk_util(client, cmd);
1686 break;
1687 }
1688 case FIO_NET_CMD_TS: {
1689 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1690
1691 dprint(FD_NET, "client: ts->ss_state = %u\n", (unsigned int) le32_to_cpu(p->ts.ss_state));
1692 if (le32_to_cpu(p->ts.ss_state) & __FIO_SS_DATA) {
1693 dprint(FD_NET, "client: received steadystate ring buffers\n");
1694
1695 size = le64_to_cpu(p->ts.ss_dur);
1696 p->ts.ss_iops_data = (uint64_t *) ((struct cmd_ts_pdu *)cmd->payload + 1);
1697 p->ts.ss_bw_data = p->ts.ss_iops_data + size;
1698 }
1699
1700 convert_ts(&p->ts, &p->ts);
1701 convert_gs(&p->rs, &p->rs);
1702
1703 ops->thread_status(client, cmd);
1704 break;
1705 }
1706 case FIO_NET_CMD_GS: {
1707 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1708
1709 convert_gs(gs, gs);
1710
1711 ops->group_stats(client, cmd);
1712 break;
1713 }
1714 case FIO_NET_CMD_ETA: {
1715 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1716
1717 if (!remove_reply_cmd(client, cmd))
1718 break;
1719 convert_jobs_eta(je);
1720 handle_eta(client, cmd);
1721 break;
1722 }
1723 case FIO_NET_CMD_PROBE:
1724 remove_reply_cmd(client, cmd);
1725 ops->probe(client, cmd);
1726 break;
1727 case FIO_NET_CMD_SERVER_START:
1728 client->state = Client_running;
1729 if (ops->job_start)
1730 ops->job_start(client, cmd);
1731 break;
1732 case FIO_NET_CMD_START: {
1733 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1734
1735 pdu->jobs = le32_to_cpu(pdu->jobs);
1736 ops->start(client, cmd);
1737 break;
1738 }
1739 case FIO_NET_CMD_STOP: {
1740 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1741
1742 convert_stop(cmd);
1743 client->state = Client_stopped;
1744 client->error = le32_to_cpu(pdu->error);
1745 client->signal = le32_to_cpu(pdu->signal);
1746 ops->stop(client, cmd);
1747 break;
1748 }
1749 case FIO_NET_CMD_ADD_JOB: {
1750 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1751
1752 client->thread_number = le32_to_cpu(pdu->thread_number);
1753 client->groupid = le32_to_cpu(pdu->groupid);
1754
1755 if (ops->add_job)
1756 ops->add_job(client, cmd);
1757 break;
1758 }
1759 case FIO_NET_CMD_IOLOG:
1760 fio_client_handle_iolog(client, cmd);
1761 break;
1762 case FIO_NET_CMD_UPDATE_JOB:
1763 ops->update_job(client, cmd);
1764 remove_reply_cmd(client, cmd);
1765 break;
1766 case FIO_NET_CMD_VTRIGGER: {
1767 struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
1768 char buf[128];
1769 int off = 0;
1770
1771 if (aux_path) {
1772 strcpy(buf, aux_path);
1773 off = strlen(buf);
1774 }
1775
1776 __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
1777 exec_trigger(trigger_cmd);
1778 break;
1779 }
1780 case FIO_NET_CMD_SENDFILE: {
1781 struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload;
1782 fio_send_file(client, pdu, cmd->tag);
1783 break;
1784 }
1785 case FIO_NET_CMD_JOB_OPT: {
1786 handle_job_opt(client, cmd);
1787 break;
1788 }
1789 default:
1790 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1791 break;
1792 }
1793
1794 free(cmd);
1795 return 1;
1796}
1797
1798int fio_clients_send_trigger(const char *cmd)
1799{
1800 struct flist_head *entry;
1801 struct fio_client *client;
1802 size_t slen;
1803
1804 dprint(FD_NET, "client: send vtrigger: %s\n", cmd);
1805
1806 if (!cmd)
1807 slen = 0;
1808 else
1809 slen = strlen(cmd);
1810
1811 flist_for_each(entry, &client_list) {
1812 struct cmd_vtrigger_pdu *pdu;
1813
1814 client = flist_entry(entry, struct fio_client, list);
1815
1816 pdu = malloc(sizeof(*pdu) + slen);
1817 pdu->len = cpu_to_le16((uint16_t) slen);
1818 if (slen)
1819 memcpy(pdu->cmd, cmd, slen);
1820 fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu,
1821 sizeof(*pdu) + slen, NULL, NULL);
1822 free(pdu);
1823 }
1824
1825 return 0;
1826}
1827
1828static void request_client_etas(struct client_ops *ops)
1829{
1830 struct fio_client *client;
1831 struct flist_head *entry;
1832 struct client_eta *eta;
1833 int skipped = 0;
1834
1835 dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1836
1837 eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
1838 eta->pending = nr_clients;
1839
1840 flist_for_each(entry, &client_list) {
1841 client = flist_entry(entry, struct fio_client, list);
1842
1843 if (!flist_empty(&client->eta_list)) {
1844 skipped++;
1845 continue;
1846 }
1847 if (client->state != Client_running)
1848 continue;
1849
1850 assert(!client->eta_in_flight);
1851 flist_add_tail(&client->eta_list, &eta_list);
1852 client->eta_in_flight = eta;
1853 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1854 (uintptr_t) eta, &client->cmd_list);
1855 }
1856
1857 while (skipped--) {
1858 if (!fio_client_dec_jobs_eta(eta, ops->eta))
1859 break;
1860 }
1861
1862 dprint(FD_NET, "client: requested eta tag %p\n", eta);
1863}
1864
1865/*
1866 * A single SEND_ETA timeout isn't fatal. Attempt to recover.
1867 */
1868static int handle_cmd_timeout(struct fio_client *client,
1869 struct fio_net_cmd_reply *reply)
1870{
1871 uint16_t reply_opcode = reply->opcode;
1872
1873 flist_del(&reply->list);
1874 free(reply);
1875
1876 if (reply_opcode != FIO_NET_CMD_SEND_ETA)
1877 return 1;
1878
1879 log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
1880
1881 flist_del_init(&client->eta_list);
1882 if (client->eta_in_flight) {
1883 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
1884 client->eta_in_flight = NULL;
1885 }
1886
1887 /*
1888 * If we fail 5 in a row, give up...
1889 */
1890 if (client->eta_timeouts++ > 5)
1891 return 1;
1892
1893 return 0;
1894}
1895
1896static int client_check_cmd_timeout(struct fio_client *client,
1897 struct timespec *now)
1898{
1899 struct fio_net_cmd_reply *reply;
1900 struct flist_head *entry, *tmp;
1901 int ret = 0;
1902
1903 flist_for_each_safe(entry, tmp, &client->cmd_list) {
1904 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1905
1906 if (mtime_since(&reply->ts, now) < FIO_NET_CLIENT_TIMEOUT)
1907 continue;
1908
1909 if (!handle_cmd_timeout(client, reply))
1910 continue;
1911
1912 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1913 fio_server_op(reply->opcode));
1914 ret = 1;
1915 }
1916
1917 return flist_empty(&client->cmd_list) && ret;
1918}
1919
1920static int fio_check_clients_timed_out(void)
1921{
1922 struct fio_client *client;
1923 struct flist_head *entry, *tmp;
1924 struct timespec ts;
1925 int ret = 0;
1926
1927 fio_gettime(&ts, NULL);
1928
1929 flist_for_each_safe(entry, tmp, &client_list) {
1930 client = flist_entry(entry, struct fio_client, list);
1931
1932 if (flist_empty(&client->cmd_list))
1933 continue;
1934
1935 if (!client_check_cmd_timeout(client, &ts))
1936 continue;
1937
1938 if (client->ops->timed_out)
1939 client->ops->timed_out(client);
1940 else
1941 log_err("fio: client %s timed out\n", client->hostname);
1942
1943 client->error = ETIMEDOUT;
1944 remove_client(client);
1945 ret = 1;
1946 }
1947
1948 return ret;
1949}
1950
1951int fio_handle_clients(struct client_ops *ops)
1952{
1953 struct pollfd *pfds;
1954 int i, ret = 0, retval = 0;
1955
1956 fio_gettime(&eta_ts, NULL);
1957
1958 pfds = malloc(nr_clients * sizeof(struct pollfd));
1959
1960 init_thread_stat(&client_ts);
1961 init_group_run_stat(&client_gs);
1962
1963 while (!exit_backend && nr_clients) {
1964 struct flist_head *entry, *tmp;
1965 struct fio_client *client;
1966
1967 i = 0;
1968 flist_for_each_safe(entry, tmp, &client_list) {
1969 client = flist_entry(entry, struct fio_client, list);
1970
1971 if (!client->sent_job && !client->ops->stay_connected &&
1972 flist_empty(&client->cmd_list)) {
1973 remove_client(client);
1974 continue;
1975 }
1976
1977 pfds[i].fd = client->fd;
1978 pfds[i].events = POLLIN;
1979 i++;
1980 }
1981
1982 if (!nr_clients)
1983 break;
1984
1985 assert(i == nr_clients);
1986
1987 do {
1988 struct timespec ts;
1989 int timeout;
1990
1991 fio_gettime(&ts, NULL);
1992 if (mtime_since(&eta_ts, &ts) >= 900) {
1993 request_client_etas(ops);
1994 memcpy(&eta_ts, &ts, sizeof(ts));
1995
1996 if (fio_check_clients_timed_out())
1997 break;
1998 }
1999
2000 check_trigger_file();
2001
2002 timeout = min(100u, ops->eta_msec);
2003
2004 ret = poll(pfds, nr_clients, timeout);
2005 if (ret < 0) {
2006 if (errno == EINTR)
2007 continue;
2008 log_err("fio: poll clients: %s\n", strerror(errno));
2009 break;
2010 } else if (!ret)
2011 continue;
2012 } while (ret <= 0);
2013
2014 for (i = 0; i < nr_clients; i++) {
2015 if (!(pfds[i].revents & POLLIN))
2016 continue;
2017
2018 client = find_client_by_fd(pfds[i].fd);
2019 if (!client) {
2020 log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd);
2021 continue;
2022 }
2023 if (!fio_handle_client(client)) {
2024 log_info("client: host=%s disconnected\n",
2025 client->hostname);
2026 remove_client(client);
2027 retval = 1;
2028 } else if (client->error)
2029 retval = 1;
2030 fio_put_client(client);
2031 }
2032 }
2033
2034 fio_client_json_fini();
2035
2036 free(pfds);
2037 return retval || error_clients;
2038}