fio: Add advise THP option to mmap engine
[fio.git] / client.c
1 #include <stdio.h>
2 #include <stdlib.h>
3 #include <unistd.h>
4 #include <errno.h>
5 #include <fcntl.h>
6 #include <poll.h>
7 #include <sys/types.h>
8 #include <sys/stat.h>
9 #include <sys/socket.h>
10 #include <sys/un.h>
11 #include <netinet/in.h>
12 #include <arpa/inet.h>
13 #include <netdb.h>
14 #include <signal.h>
15 #ifdef CONFIG_ZLIB
16 #include <zlib.h>
17 #endif
18
19 #include "fio.h"
20 #include "client.h"
21 #include "server.h"
22 #include "flist.h"
23 #include "hash.h"
24 #include "verify-state.h"
25
26 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
27 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd);
28 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd);
29 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
30 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd);
31 static void handle_stop(struct fio_client *client);
32 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd);
33
34 static void convert_text(struct fio_net_cmd *cmd);
35 static void client_display_thread_status(struct jobs_eta *je);
36
37 struct client_ops fio_client_ops = {
38         .text           = handle_text,
39         .disk_util      = handle_du,
40         .thread_status  = handle_ts,
41         .group_stats    = handle_gs,
42         .stop           = handle_stop,
43         .start          = handle_start,
44         .eta            = client_display_thread_status,
45         .probe          = handle_probe,
46         .eta_msec       = FIO_CLIENT_DEF_ETA_MSEC,
47         .client_type    = FIO_CLIENT_TYPE_CLI,
48 };
49
50 static struct timespec eta_ts;
51
52 static FLIST_HEAD(client_list);
53 static FLIST_HEAD(eta_list);
54
55 static FLIST_HEAD(arg_list);
56
57 struct thread_stat client_ts;
58 struct group_run_stats client_gs;
59 int sum_stat_clients;
60
61 static int sum_stat_nr;
62 static struct buf_output allclients;
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
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
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
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
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
123 static int read_ini_data(int fd, void *data, size_t size)
124 {
125         char *p = data;
126         int ret = 0;
127         FILE *fp;
128         int dupfd;
129
130         dupfd = dup(fd);
131         if (dupfd < 0)
132                 return errno;
133
134         fp = fdopen(dupfd, "r");
135         if (!fp) {
136                 ret = errno;
137                 close(dupfd);
138                 goto out;
139         }
140
141         while (1) {
142                 ssize_t len;
143                 char buf[OPT_LEN_MAX+1], *sub;
144
145                 if (!fgets(buf, sizeof(buf), fp)) {
146                         if (ferror(fp)) {
147                                 if (errno == EAGAIN || errno == EINTR)
148                                         continue;
149                                 ret = errno;
150                         }
151                         break;
152                 }
153
154                 sub = fio_option_dup_subs(buf);
155                 len = strlen(sub);
156                 if (len + 1 > size) {
157                         log_err("fio: no space left to read data\n");
158                         free(sub);
159                         ret = ENOSPC;
160                         break;
161                 }
162
163                 memcpy(p, sub, len);
164                 free(sub);
165                 p += len;
166                 *p = '\0';
167                 size -= len;
168         }
169
170         fclose(fp);
171 out:
172         return ret;
173 }
174
175 static void fio_client_json_init(void)
176 {
177         char time_buf[32];
178         time_t time_p;
179
180         if (!(output_format & FIO_OUTPUT_JSON))
181                 return;
182
183         time(&time_p);
184         os_ctime_r((const time_t *) &time_p, time_buf, sizeof(time_buf));
185         time_buf[strlen(time_buf) - 1] = '\0';
186
187         root = json_create_object();
188         json_object_add_value_string(root, "fio version", fio_version_string);
189         json_object_add_value_int(root, "timestamp", time_p);
190         json_object_add_value_string(root, "time", time_buf);
191
192         job_opt_object = json_create_object();
193         json_object_add_value_object(root, "global options", job_opt_object);
194         clients_array = json_create_array();
195         json_object_add_value_array(root, "client_stats", clients_array);
196         du_array = json_create_array();
197         json_object_add_value_array(root, "disk_util", du_array);
198 }
199
200 static void fio_client_json_fini(void)
201 {
202         struct buf_output out;
203
204         if (!root)
205                 return;
206
207         buf_output_init(&out);
208
209         __log_buf(&out, "\n");
210         json_print_object(root, &out);
211         __log_buf(&out, "\n");
212         log_info_buf(out.buf, out.buflen);
213
214         buf_output_free(&out);
215
216         json_free_object(root);
217         root = NULL;
218         job_opt_object = NULL;
219         clients_array = NULL;
220         du_array = NULL;
221 }
222
223 static struct fio_client *find_client_by_fd(int fd)
224 {
225         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
226         struct fio_client *client;
227         struct flist_head *entry;
228
229         flist_for_each(entry, &client_hash[bucket]) {
230                 client = flist_entry(entry, struct fio_client, hash_list);
231
232                 if (client->fd == fd) {
233                         client->refs++;
234                         return client;
235                 }
236         }
237
238         return NULL;
239 }
240
241 void fio_put_client(struct fio_client *client)
242 {
243         if (--client->refs)
244                 return;
245
246         log_info_buf(client->buf.buf, client->buf.buflen);
247         buf_output_free(&client->buf);
248
249         free(client->hostname);
250         if (client->argv)
251                 free(client->argv);
252         if (client->name)
253                 free(client->name);
254         while (client->nr_files) {
255                 struct client_file *cf = &client->files[--client->nr_files];
256
257                 free(cf->file);
258         }
259         if (client->files)
260                 free(client->files);
261         if (client->opt_lists)
262                 free(client->opt_lists);
263
264         if (!client->did_stat)
265                 sum_stat_clients--;
266
267         if (client->error)
268                 error_clients++;
269
270         free(client);
271 }
272
273 static int fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
274 {
275         if (!--eta->pending) {
276                 eta_fn(&eta->eta);
277                 free(eta);
278                 return 0;
279         }
280
281         return 1;
282 }
283
284 static void fio_drain_client_text(struct fio_client *client)
285 {
286         do {
287                 struct fio_net_cmd *cmd;
288
289                 cmd = fio_net_recv_cmd(client->fd, false);
290                 if (!cmd)
291                         break;
292
293                 if (cmd->opcode == FIO_NET_CMD_TEXT) {
294                         convert_text(cmd);
295                         client->ops->text(client, cmd);
296                 }
297
298                 free(cmd);
299         } while (1);
300 }
301
302 static void remove_client(struct fio_client *client)
303 {
304         assert(client->refs);
305
306         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
307
308         fio_drain_client_text(client);
309
310         if (!flist_empty(&client->list))
311                 flist_del_init(&client->list);
312
313         fio_client_remove_hash(client);
314
315         if (!flist_empty(&client->eta_list)) {
316                 flist_del_init(&client->eta_list);
317                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
318         }
319
320         close(client->fd);
321         client->fd = -1;
322
323         if (client->ops->removed)
324                 client->ops->removed(client);
325
326         nr_clients--;
327         fio_put_client(client);
328 }
329
330 struct fio_client *fio_get_client(struct fio_client *client)
331 {
332         client->refs++;
333         return client;
334 }
335
336 static void __fio_client_add_cmd_option(struct fio_client *client,
337                                         const char *opt)
338 {
339         int index;
340
341         index = client->argc++;
342         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
343         client->argv[index] = strdup(opt);
344         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
345 }
346
347 void fio_client_add_cmd_option(void *cookie, const char *opt)
348 {
349         struct fio_client *client = cookie;
350         struct flist_head *entry;
351
352         if (!client || !opt)
353                 return;
354
355         __fio_client_add_cmd_option(client, opt);
356
357         /*
358          * Duplicate arguments to shared client group
359          */
360         flist_for_each(entry, &arg_list) {
361                 client = flist_entry(entry, struct fio_client, arg_list);
362
363                 __fio_client_add_cmd_option(client, opt);
364         }
365 }
366
367 static struct fio_client *get_new_client(void)
368 {
369         struct fio_client *client;
370
371         client = malloc(sizeof(*client));
372         memset(client, 0, sizeof(*client));
373
374         INIT_FLIST_HEAD(&client->list);
375         INIT_FLIST_HEAD(&client->hash_list);
376         INIT_FLIST_HEAD(&client->arg_list);
377         INIT_FLIST_HEAD(&client->eta_list);
378         INIT_FLIST_HEAD(&client->cmd_list);
379
380         buf_output_init(&client->buf);
381
382         return client;
383 }
384
385 struct fio_client *fio_client_add_explicit(struct client_ops *ops,
386                                            const char *hostname, int type,
387                                            int port)
388 {
389         struct fio_client *client;
390
391         client = get_new_client();
392
393         client->hostname = strdup(hostname);
394
395         if (type == Fio_client_socket)
396                 client->is_sock = true;
397         else {
398                 int ipv6;
399
400                 ipv6 = type == Fio_client_ipv6;
401                 if (fio_server_parse_host(hostname, ipv6,
402                                                 &client->addr.sin_addr,
403                                                 &client->addr6.sin6_addr))
404                         goto err;
405
406                 client->port = port;
407         }
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         return client;
420 err:
421         free(client);
422         return NULL;
423 }
424
425 int fio_client_add_ini_file(void *cookie, const char *ini_file, bool remote)
426 {
427         struct fio_client *client = cookie;
428         struct client_file *cf;
429         size_t new_size;
430         void *new_files;
431
432         if (!client)
433                 return 1;
434
435         dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file);
436
437         new_size = (client->nr_files + 1) * sizeof(struct client_file);
438         new_files = realloc(client->files, new_size);
439         if (!new_files)
440                 return 1;
441
442         client->files = new_files;
443         cf = &client->files[client->nr_files];
444         cf->file = strdup(ini_file);
445         cf->remote = remote;
446         client->nr_files++;
447         return 0;
448 }
449
450 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
451 {
452         struct fio_client *existing = *cookie;
453         struct fio_client *client;
454
455         if (existing) {
456                 /*
457                  * We always add our "exec" name as the option, hence 1
458                  * means empty.
459                  */
460                 if (existing->argc == 1)
461                         flist_add_tail(&existing->arg_list, &arg_list);
462                 else {
463                         while (!flist_empty(&arg_list))
464                                 flist_del_init(arg_list.next);
465                 }
466         }
467
468         client = get_new_client();
469
470         if (fio_server_parse_string(hostname, &client->hostname,
471                                         &client->is_sock, &client->port,
472                                         &client->addr.sin_addr,
473                                         &client->addr6.sin6_addr,
474                                         &client->ipv6))
475                 return -1;
476
477         client->fd = -1;
478         client->ops = ops;
479         client->refs = 1;
480         client->type = ops->client_type;
481
482         __fio_client_add_cmd_option(client, "fio");
483
484         flist_add(&client->list, &client_list);
485         nr_clients++;
486         dprint(FD_NET, "client: added <%s>\n", client->hostname);
487         *cookie = client;
488         return 0;
489 }
490
491 static const char *server_name(struct fio_client *client, char *buf,
492                                size_t bufsize)
493 {
494         const char *from;
495
496         if (client->ipv6)
497                 from = inet_ntop(AF_INET6, (struct sockaddr *) &client->addr6.sin6_addr, buf, bufsize);
498         else if (client->is_sock)
499                 from = "sock";
500         else
501                 from = inet_ntop(AF_INET, (struct sockaddr *) &client->addr.sin_addr, buf, bufsize);
502
503         return from;
504 }
505
506 static void probe_client(struct fio_client *client)
507 {
508         struct cmd_client_probe_pdu pdu;
509         const char *sname;
510         uint64_t tag;
511         char buf[64];
512
513         dprint(FD_NET, "client: send probe\n");
514
515 #ifdef CONFIG_ZLIB
516         pdu.flags = __le64_to_cpu(FIO_PROBE_FLAG_ZLIB);
517 #else
518         pdu.flags = 0;
519 #endif
520
521         sname = server_name(client, buf, sizeof(buf));
522         memset(pdu.server, 0, sizeof(pdu.server));
523         strncpy((char *) pdu.server, sname, sizeof(pdu.server) - 1);
524
525         fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list);
526 }
527
528 static int fio_client_connect_ip(struct fio_client *client)
529 {
530         struct sockaddr *addr;
531         socklen_t socklen;
532         int fd, domain;
533
534         if (client->ipv6) {
535                 client->addr6.sin6_family = AF_INET6;
536                 client->addr6.sin6_port = htons(client->port);
537                 domain = AF_INET6;
538                 addr = (struct sockaddr *) &client->addr6;
539                 socklen = sizeof(client->addr6);
540         } else {
541                 client->addr.sin_family = AF_INET;
542                 client->addr.sin_port = htons(client->port);
543                 domain = AF_INET;
544                 addr = (struct sockaddr *) &client->addr;
545                 socklen = sizeof(client->addr);
546         }
547
548         fd = socket(domain, SOCK_STREAM, 0);
549         if (fd < 0) {
550                 int ret = -errno;
551
552                 log_err("fio: socket: %s\n", strerror(errno));
553                 return ret;
554         }
555
556         if (connect(fd, addr, socklen) < 0) {
557                 int ret = -errno;
558
559                 log_err("fio: connect: %s\n", strerror(errno));
560                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
561                                                                 client->port);
562                 close(fd);
563                 return ret;
564         }
565
566         return fd;
567 }
568
569 static int fio_client_connect_sock(struct fio_client *client)
570 {
571         struct sockaddr_un *addr = &client->addr_un;
572         socklen_t len;
573         int fd;
574
575         memset(addr, 0, sizeof(*addr));
576         addr->sun_family = AF_UNIX;
577         strncpy(addr->sun_path, client->hostname, sizeof(addr->sun_path) - 1);
578
579         fd = socket(AF_UNIX, SOCK_STREAM, 0);
580         if (fd < 0) {
581                 int ret = -errno;
582
583                 log_err("fio: socket: %s\n", strerror(errno));
584                 return ret;
585         }
586
587         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
588         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
589                 int ret = -errno;
590
591                 log_err("fio: connect; %s\n", strerror(errno));
592                 close(fd);
593                 return ret;
594         }
595
596         return fd;
597 }
598
599 int fio_client_connect(struct fio_client *client)
600 {
601         int fd;
602
603         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
604
605         if (client->is_sock)
606                 fd = fio_client_connect_sock(client);
607         else
608                 fd = fio_client_connect_ip(client);
609
610         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
611
612         if (fd < 0)
613                 return fd;
614
615         client->fd = fd;
616         fio_client_add_hash(client);
617         client->state = Client_connected;
618
619         probe_client(client);
620         return 0;
621 }
622
623 int fio_client_terminate(struct fio_client *client)
624 {
625         return fio_net_send_quit(client->fd);
626 }
627
628 static void fio_clients_terminate(void)
629 {
630         struct flist_head *entry;
631         struct fio_client *client;
632
633         dprint(FD_NET, "client: terminate clients\n");
634
635         flist_for_each(entry, &client_list) {
636                 client = flist_entry(entry, struct fio_client, list);
637                 fio_client_terminate(client);
638         }
639 }
640
641 static void sig_int(int sig)
642 {
643         dprint(FD_NET, "client: got signal %d\n", sig);
644         fio_clients_terminate();
645 }
646
647 static void client_signal_handler(void)
648 {
649         struct sigaction act;
650
651         memset(&act, 0, sizeof(act));
652         act.sa_handler = sig_int;
653         act.sa_flags = SA_RESTART;
654         sigaction(SIGINT, &act, NULL);
655
656         memset(&act, 0, sizeof(act));
657         act.sa_handler = sig_int;
658         act.sa_flags = SA_RESTART;
659         sigaction(SIGTERM, &act, NULL);
660
661 /* Windows uses SIGBREAK as a quit signal from other applications */
662 #ifdef WIN32
663         memset(&act, 0, sizeof(act));
664         act.sa_handler = sig_int;
665         act.sa_flags = SA_RESTART;
666         sigaction(SIGBREAK, &act, NULL);
667 #endif
668
669         memset(&act, 0, sizeof(act));
670         act.sa_handler = sig_show_status;
671         act.sa_flags = SA_RESTART;
672         sigaction(SIGUSR1, &act, NULL);
673 }
674
675 static int send_client_cmd_line(struct fio_client *client)
676 {
677         struct cmd_single_line_pdu *cslp;
678         struct cmd_line_pdu *clp;
679         unsigned long offset;
680         unsigned int *lens;
681         void *pdu;
682         size_t mem;
683         int i, ret;
684
685         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
686
687         lens = malloc(client->argc * sizeof(unsigned int));
688
689         /*
690          * Find out how much mem we need
691          */
692         for (i = 0, mem = 0; i < client->argc; i++) {
693                 lens[i] = strlen(client->argv[i]) + 1;
694                 mem += lens[i];
695         }
696
697         /*
698          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
699          */
700         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
701
702         pdu = malloc(mem);
703         clp = pdu;
704         offset = sizeof(*clp);
705
706         for (i = 0; i < client->argc; i++) {
707                 uint16_t arg_len = lens[i];
708
709                 cslp = pdu + offset;
710                 strcpy((char *) cslp->text, client->argv[i]);
711                 cslp->len = cpu_to_le16(arg_len);
712                 offset += sizeof(*cslp) + arg_len;
713         }
714
715         free(lens);
716         clp->lines = cpu_to_le16(client->argc);
717         clp->client_type = __cpu_to_le16(client->type);
718         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, NULL, NULL);
719         free(pdu);
720         return ret;
721 }
722
723 int fio_clients_connect(void)
724 {
725         struct fio_client *client;
726         struct flist_head *entry, *tmp;
727         int ret;
728
729 #ifdef WIN32
730         WSADATA wsd;
731         WSAStartup(MAKEWORD(2, 2), &wsd);
732 #endif
733
734         dprint(FD_NET, "client: connect all\n");
735
736         client_signal_handler();
737
738         flist_for_each_safe(entry, tmp, &client_list) {
739                 client = flist_entry(entry, struct fio_client, list);
740
741                 ret = fio_client_connect(client);
742                 if (ret) {
743                         remove_client(client);
744                         continue;
745                 }
746
747                 if (client->argc > 1)
748                         send_client_cmd_line(client);
749         }
750
751         return !nr_clients;
752 }
753
754 int fio_start_client(struct fio_client *client)
755 {
756         dprint(FD_NET, "client: start %s\n", client->hostname);
757         return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
758 }
759
760 int fio_start_all_clients(void)
761 {
762         struct fio_client *client;
763         struct flist_head *entry, *tmp;
764         int ret;
765
766         dprint(FD_NET, "client: start all\n");
767
768         fio_client_json_init();
769
770         flist_for_each_safe(entry, tmp, &client_list) {
771                 client = flist_entry(entry, struct fio_client, list);
772
773                 ret = fio_start_client(client);
774                 if (ret) {
775                         remove_client(client);
776                         continue;
777                 }
778         }
779
780         return flist_empty(&client_list);
781 }
782
783 static int __fio_client_send_remote_ini(struct fio_client *client,
784                                         const char *filename)
785 {
786         struct cmd_load_file_pdu *pdu;
787         size_t p_size;
788         int ret;
789
790         dprint(FD_NET, "send remote ini %s to %s\n", filename, client->hostname);
791
792         p_size = sizeof(*pdu) + strlen(filename) + 1;
793         pdu = malloc(p_size);
794         memset(pdu, 0, p_size);
795         pdu->name_len = strlen(filename);
796         strcpy((char *) pdu->file, filename);
797         pdu->client_type = cpu_to_le16((uint16_t) client->type);
798
799         client->sent_job = true;
800         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_LOAD_FILE, pdu, p_size,NULL, NULL);
801         free(pdu);
802         return ret;
803 }
804
805 /*
806  * Send file contents to server backend. We could use sendfile(), but to remain
807  * more portable lets just read/write the darn thing.
808  */
809 static int __fio_client_send_local_ini(struct fio_client *client,
810                                        const char *filename)
811 {
812         struct cmd_job_pdu *pdu;
813         size_t p_size;
814         struct stat sb;
815         char *p;
816         void *buf;
817         off_t len;
818         int fd, ret;
819
820         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
821
822         fd = open(filename, O_RDONLY);
823         if (fd < 0) {
824                 ret = -errno;
825                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
826                 return ret;
827         }
828
829         if (fstat(fd, &sb) < 0) {
830                 ret = -errno;
831                 log_err("fio: job file stat: %s\n", strerror(errno));
832                 close(fd);
833                 return ret;
834         }
835
836         /*
837          * Add extra space for variable expansion, but doesn't guarantee.
838          */
839         sb.st_size += OPT_LEN_MAX;
840         p_size = sb.st_size + sizeof(*pdu);
841         pdu = malloc(p_size);
842         buf = pdu->buf;
843
844         len = sb.st_size;
845         p = buf;
846         if (read_ini_data(fd, p, len)) {
847                 log_err("fio: failed reading job file %s\n", filename);
848                 close(fd);
849                 free(pdu);
850                 return 1;
851         }
852
853         pdu->buf_len = __cpu_to_le32(sb.st_size);
854         pdu->client_type = cpu_to_le32(client->type);
855
856         client->sent_job = true;
857         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, NULL, NULL);
858         free(pdu);
859         close(fd);
860         return ret;
861 }
862
863 int fio_client_send_ini(struct fio_client *client, const char *filename,
864                         bool remote)
865 {
866         int ret;
867
868         if (!remote)
869                 ret = __fio_client_send_local_ini(client, filename);
870         else
871                 ret = __fio_client_send_remote_ini(client, filename);
872
873         if (!ret)
874                 client->sent_job = true;
875
876         return ret;
877 }
878
879 static int fio_client_send_cf(struct fio_client *client,
880                               struct client_file *cf)
881 {
882         return fio_client_send_ini(client, cf->file, cf->remote);
883 }
884
885 int fio_clients_send_ini(const char *filename)
886 {
887         struct fio_client *client;
888         struct flist_head *entry, *tmp;
889
890         flist_for_each_safe(entry, tmp, &client_list) {
891                 bool failed = false;
892
893                 client = flist_entry(entry, struct fio_client, list);
894
895                 if (client->nr_files) {
896                         int i;
897
898                         for (i = 0; i < client->nr_files; i++) {
899                                 struct client_file *cf;
900
901                                 cf = &client->files[i];
902
903                                 if (fio_client_send_cf(client, cf)) {
904                                         failed = true;
905                                         remove_client(client);
906                                         break;
907                                 }
908                         }
909                 }
910                 if (client->sent_job || failed)
911                         continue;
912                 if (!filename || fio_client_send_ini(client, filename, 0))
913                         remove_client(client);
914         }
915
916         return !nr_clients;
917 }
918
919 int fio_client_update_options(struct fio_client *client,
920                               struct thread_options *o, uint64_t *tag)
921 {
922         struct cmd_add_job_pdu pdu;
923
924         pdu.thread_number = cpu_to_le32(client->thread_number);
925         pdu.groupid = cpu_to_le32(client->groupid);
926         convert_thread_options_to_net(&pdu.top, o);
927
928         return fio_net_send_cmd(client->fd, FIO_NET_CMD_UPDATE_JOB, &pdu, sizeof(pdu), tag, &client->cmd_list);
929 }
930
931 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
932 {
933         dst->max_val    = le64_to_cpu(src->max_val);
934         dst->min_val    = le64_to_cpu(src->min_val);
935         dst->samples    = le64_to_cpu(src->samples);
936
937         /*
938          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
939          */
940         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
941         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
942 }
943
944 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
945 {
946         int i, j;
947
948         dst->error              = le32_to_cpu(src->error);
949         dst->thread_number      = le32_to_cpu(src->thread_number);
950         dst->groupid            = le32_to_cpu(src->groupid);
951         dst->pid                = le32_to_cpu(src->pid);
952         dst->members            = le32_to_cpu(src->members);
953         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
954
955         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
956                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
957                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
958                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
959                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
960                 convert_io_stat(&dst->iops_stat[i], &src->iops_stat[i]);
961         }
962
963         dst->usr_time           = le64_to_cpu(src->usr_time);
964         dst->sys_time           = le64_to_cpu(src->sys_time);
965         dst->ctx                = le64_to_cpu(src->ctx);
966         dst->minf               = le64_to_cpu(src->minf);
967         dst->majf               = le64_to_cpu(src->majf);
968         dst->clat_percentiles   = le32_to_cpu(src->clat_percentiles);
969         dst->lat_percentiles    = le32_to_cpu(src->lat_percentiles);
970         dst->percentile_precision = le64_to_cpu(src->percentile_precision);
971
972         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
973                 fio_fp64_t *fps = &src->percentile_list[i];
974                 fio_fp64_t *fpd = &dst->percentile_list[i];
975
976                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
977         }
978
979         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
980                 dst->io_u_map[i]        = le64_to_cpu(src->io_u_map[i]);
981                 dst->io_u_submit[i]     = le64_to_cpu(src->io_u_submit[i]);
982                 dst->io_u_complete[i]   = le64_to_cpu(src->io_u_complete[i]);
983         }
984
985         for (i = 0; i < FIO_IO_U_LAT_N_NR; i++)
986                 dst->io_u_lat_n[i]      = le64_to_cpu(src->io_u_lat_n[i]);
987         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
988                 dst->io_u_lat_u[i]      = le64_to_cpu(src->io_u_lat_u[i]);
989         for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
990                 dst->io_u_lat_m[i]      = le64_to_cpu(src->io_u_lat_m[i]);
991
992         for (i = 0; i < DDIR_RWDIR_CNT; i++)
993                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
994                         dst->io_u_plat[i][j] = le64_to_cpu(src->io_u_plat[i][j]);
995
996         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
997                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
998                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
999                 dst->drop_io_u[i]       = le64_to_cpu(src->drop_io_u[i]);
1000         }
1001
1002         dst->total_submit       = le64_to_cpu(src->total_submit);
1003         dst->total_complete     = le64_to_cpu(src->total_complete);
1004         dst->nr_zone_resets     = le64_to_cpu(src->nr_zone_resets);
1005
1006         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1007                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
1008                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
1009         }
1010
1011         dst->total_run_time     = le64_to_cpu(src->total_run_time);
1012         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
1013         dst->total_err_count    = le64_to_cpu(src->total_err_count);
1014         dst->first_error        = le32_to_cpu(src->first_error);
1015         dst->kb_base            = le32_to_cpu(src->kb_base);
1016         dst->unit_base          = le32_to_cpu(src->unit_base);
1017
1018         dst->sig_figs           = le32_to_cpu(src->sig_figs);
1019
1020         dst->latency_depth      = le32_to_cpu(src->latency_depth);
1021         dst->latency_target     = le64_to_cpu(src->latency_target);
1022         dst->latency_window     = le64_to_cpu(src->latency_window);
1023         dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
1024
1025         dst->nr_block_infos     = le64_to_cpu(src->nr_block_infos);
1026         for (i = 0; i < dst->nr_block_infos; i++)
1027                 dst->block_infos[i] = le32_to_cpu(src->block_infos[i]);
1028
1029         dst->ss_dur             = le64_to_cpu(src->ss_dur);
1030         dst->ss_state           = le32_to_cpu(src->ss_state);
1031         dst->ss_head            = le32_to_cpu(src->ss_head);
1032         dst->ss_limit.u.f       = fio_uint64_to_double(le64_to_cpu(src->ss_limit.u.i));
1033         dst->ss_slope.u.f       = fio_uint64_to_double(le64_to_cpu(src->ss_slope.u.i));
1034         dst->ss_deviation.u.f   = fio_uint64_to_double(le64_to_cpu(src->ss_deviation.u.i));
1035         dst->ss_criterion.u.f   = fio_uint64_to_double(le64_to_cpu(src->ss_criterion.u.i));
1036
1037         if (dst->ss_state & FIO_SS_DATA) {
1038                 for (i = 0; i < dst->ss_dur; i++ ) {
1039                         dst->ss_iops_data[i] = le64_to_cpu(src->ss_iops_data[i]);
1040                         dst->ss_bw_data[i] = le64_to_cpu(src->ss_bw_data[i]);
1041                 }
1042         }
1043
1044         dst->cachehit           = le64_to_cpu(src->cachehit);
1045         dst->cachemiss          = le64_to_cpu(src->cachemiss);
1046 }
1047
1048 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
1049 {
1050         int i;
1051
1052         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1053                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
1054                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
1055                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
1056                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
1057                 dst->iobytes[i]         = le64_to_cpu(src->iobytes[i]);
1058                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
1059         }
1060
1061         dst->kb_base    = le32_to_cpu(src->kb_base);
1062         dst->unit_base  = le32_to_cpu(src->unit_base);
1063         dst->sig_figs   = le32_to_cpu(src->sig_figs);
1064         dst->groupid    = le32_to_cpu(src->groupid);
1065         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
1066 }
1067
1068 static void json_object_add_client_info(struct json_object *obj,
1069                                         struct fio_client *client)
1070 {
1071         const char *hostname = client->hostname ? client->hostname : "";
1072
1073         json_object_add_value_string(obj, "hostname", hostname);
1074         json_object_add_value_int(obj, "port", client->port);
1075 }
1076
1077 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
1078 {
1079         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1080         struct flist_head *opt_list = NULL;
1081         struct json_object *tsobj;
1082
1083         if (client->opt_lists && p->ts.thread_number <= client->jobs)
1084                 opt_list = &client->opt_lists[p->ts.thread_number - 1];
1085
1086         tsobj = show_thread_status(&p->ts, &p->rs, opt_list, &client->buf);
1087         client->did_stat = true;
1088         if (tsobj) {
1089                 json_object_add_client_info(tsobj, client);
1090                 json_array_add_value_object(clients_array, tsobj);
1091         }
1092
1093         if (sum_stat_clients <= 1)
1094                 return;
1095
1096         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1);
1097         sum_group_stats(&client_gs, &p->rs);
1098
1099         client_ts.members++;
1100         client_ts.thread_number = p->ts.thread_number;
1101         client_ts.groupid = p->ts.groupid;
1102         client_ts.unified_rw_rep = p->ts.unified_rw_rep;
1103         client_ts.sig_figs = p->ts.sig_figs;
1104
1105         if (++sum_stat_nr == sum_stat_clients) {
1106                 strcpy(client_ts.name, "All clients");
1107                 tsobj = show_thread_status(&client_ts, &client_gs, NULL, &allclients);
1108                 if (tsobj) {
1109                         json_object_add_client_info(tsobj, client);
1110                         json_array_add_value_object(clients_array, tsobj);
1111                 }
1112         }
1113 }
1114
1115 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
1116 {
1117         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1118
1119         if (output_format & FIO_OUTPUT_NORMAL)
1120                 show_group_stats(gs, &client->buf);
1121 }
1122
1123 static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
1124 {
1125         struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
1126         struct print_option *p;
1127
1128         if (!job_opt_object)
1129                 return;
1130
1131         pdu->global = le16_to_cpu(pdu->global);
1132         pdu->truncated = le16_to_cpu(pdu->truncated);
1133         pdu->groupid = le32_to_cpu(pdu->groupid);
1134
1135         p = malloc(sizeof(*p));
1136         p->name = strdup((char *) pdu->name);
1137         if (pdu->value[0] != '\0')
1138                 p->value = strdup((char *) pdu->value);
1139         else
1140                 p->value = NULL;
1141
1142         if (pdu->global) {
1143                 const char *pos = "";
1144
1145                 if (p->value)
1146                         pos = p->value;
1147
1148                 json_object_add_value_string(job_opt_object, p->name, pos);
1149         } else if (client->opt_lists) {
1150                 struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
1151
1152                 flist_add_tail(&p->list, opt_list);
1153         }
1154 }
1155
1156 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
1157 {
1158         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1159         const char *buf = (const char *) pdu->buf;
1160         const char *name;
1161         int fio_unused ret;
1162         struct buf_output out;
1163
1164         buf_output_init(&out);
1165
1166         name = client->name ? client->name : client->hostname;
1167
1168         if (!client->skip_newline && !(output_format & FIO_OUTPUT_TERSE))
1169                 __log_buf(&out, "<%s> ", name);
1170         __log_buf(&out, "%s", buf);
1171         log_info_buf(out.buf, out.buflen);
1172         buf_output_free(&out);
1173         client->skip_newline = strchr(buf, '\n') == NULL;
1174 }
1175
1176 static void convert_agg(struct disk_util_agg *agg)
1177 {
1178         int i;
1179
1180         for (i = 0; i < 2; i++) {
1181                 agg->ios[i]     = le64_to_cpu(agg->ios[i]);
1182                 agg->merges[i]  = le64_to_cpu(agg->merges[i]);
1183                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
1184                 agg->ticks[i]   = le64_to_cpu(agg->ticks[i]);
1185         }
1186
1187         agg->io_ticks           = le64_to_cpu(agg->io_ticks);
1188         agg->time_in_queue      = le64_to_cpu(agg->time_in_queue);
1189         agg->slavecount         = le32_to_cpu(agg->slavecount);
1190         agg->max_util.u.f       = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
1191 }
1192
1193 static void convert_dus(struct disk_util_stat *dus)
1194 {
1195         int i;
1196
1197         for (i = 0; i < 2; i++) {
1198                 dus->s.ios[i]           = le64_to_cpu(dus->s.ios[i]);
1199                 dus->s.merges[i]        = le64_to_cpu(dus->s.merges[i]);
1200                 dus->s.sectors[i]       = le64_to_cpu(dus->s.sectors[i]);
1201                 dus->s.ticks[i]         = le64_to_cpu(dus->s.ticks[i]);
1202         }
1203
1204         dus->s.io_ticks         = le64_to_cpu(dus->s.io_ticks);
1205         dus->s.time_in_queue    = le64_to_cpu(dus->s.time_in_queue);
1206         dus->s.msec             = le64_to_cpu(dus->s.msec);
1207 }
1208
1209 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
1210 {
1211         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1212
1213         if (!client->disk_stats_shown)
1214                 client->disk_stats_shown = true;
1215
1216         if (output_format & FIO_OUTPUT_JSON) {
1217                 struct json_object *duobj;
1218
1219                 json_array_add_disk_util(&du->dus, &du->agg, du_array);
1220                 duobj = json_array_last_value_object(du_array);
1221                 json_object_add_client_info(duobj, client);
1222         } else if (output_format & FIO_OUTPUT_TERSE)
1223                 print_disk_util(&du->dus, &du->agg, 1, &client->buf);
1224         else if (output_format & FIO_OUTPUT_NORMAL) {
1225                 __log_buf(&client->buf, "\nDisk stats (read/write):\n");
1226                 print_disk_util(&du->dus, &du->agg, 0, &client->buf);
1227         }
1228 }
1229
1230 static void convert_jobs_eta(struct jobs_eta *je)
1231 {
1232         int i;
1233
1234         je->nr_running          = le32_to_cpu(je->nr_running);
1235         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
1236         je->nr_pending          = le32_to_cpu(je->nr_pending);
1237         je->nr_setting_up       = le32_to_cpu(je->nr_setting_up);
1238         je->files_open          = le32_to_cpu(je->files_open);
1239
1240         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1241                 je->m_rate[i]   = le64_to_cpu(je->m_rate[i]);
1242                 je->t_rate[i]   = le64_to_cpu(je->t_rate[i]);
1243                 je->m_iops[i]   = le32_to_cpu(je->m_iops[i]);
1244                 je->t_iops[i]   = le32_to_cpu(je->t_iops[i]);
1245                 je->rate[i]     = le64_to_cpu(je->rate[i]);
1246                 je->iops[i]     = le32_to_cpu(je->iops[i]);
1247         }
1248
1249         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
1250         je->eta_sec             = le64_to_cpu(je->eta_sec);
1251         je->nr_threads          = le32_to_cpu(je->nr_threads);
1252         je->is_pow2             = le32_to_cpu(je->is_pow2);
1253         je->unit_base           = le32_to_cpu(je->unit_base);
1254         je->sig_figs            = le32_to_cpu(je->sig_figs);
1255 }
1256
1257 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
1258 {
1259         int i;
1260
1261         dst->nr_running         += je->nr_running;
1262         dst->nr_ramp            += je->nr_ramp;
1263         dst->nr_pending         += je->nr_pending;
1264         dst->nr_setting_up      += je->nr_setting_up;
1265         dst->files_open         += je->files_open;
1266
1267         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1268                 dst->m_rate[i]  += je->m_rate[i];
1269                 dst->t_rate[i]  += je->t_rate[i];
1270                 dst->m_iops[i]  += je->m_iops[i];
1271                 dst->t_iops[i]  += je->t_iops[i];
1272                 dst->rate[i]    += je->rate[i];
1273                 dst->iops[i]    += je->iops[i];
1274         }
1275
1276         dst->elapsed_sec        += je->elapsed_sec;
1277
1278         if (je->eta_sec > dst->eta_sec)
1279                 dst->eta_sec = je->eta_sec;
1280
1281         dst->nr_threads         += je->nr_threads;
1282
1283         /*
1284          * This wont be correct for multiple strings, but at least it
1285          * works for the basic cases.
1286          */
1287         strcpy((char *) dst->run_str, (char *) je->run_str);
1288 }
1289
1290 static bool remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
1291 {
1292         struct fio_net_cmd_reply *reply = NULL;
1293         struct flist_head *entry;
1294
1295         flist_for_each(entry, &client->cmd_list) {
1296                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1297
1298                 if (cmd->tag == (uintptr_t) reply)
1299                         break;
1300
1301                 reply = NULL;
1302         }
1303
1304         if (!reply) {
1305                 log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag);
1306                 return false;
1307         }
1308
1309         flist_del(&reply->list);
1310         cmd->tag = reply->saved_tag;
1311         free(reply);
1312         return true;
1313 }
1314
1315 int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
1316 {
1317         do {
1318                 struct fio_net_cmd_reply *reply = NULL;
1319                 struct flist_head *entry;
1320
1321                 flist_for_each(entry, &client->cmd_list) {
1322                         reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1323
1324                         if (tag == (uintptr_t) reply)
1325                                 break;
1326
1327                         reply = NULL;
1328                 }
1329
1330                 if (!reply)
1331                         break;
1332
1333                 usleep(1000);
1334         } while (1);
1335
1336         return 0;
1337 }
1338
1339 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
1340 {
1341         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1342         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
1343
1344         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
1345
1346         assert(client->eta_in_flight == eta);
1347
1348         client->eta_in_flight = NULL;
1349         flist_del_init(&client->eta_list);
1350         client->eta_timeouts = 0;
1351
1352         if (client->ops->jobs_eta)
1353                 client->ops->jobs_eta(client, je);
1354
1355         fio_client_sum_jobs_eta(&eta->eta, je);
1356         fio_client_dec_jobs_eta(eta, client->ops->eta);
1357 }
1358
1359 static void client_flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
1360                                       uint64_t sample_size)
1361 {
1362         struct io_sample *s;
1363         int log_offset;
1364         uint64_t i, j, nr_samples;
1365         struct io_u_plat_entry *entry;
1366         uint64_t *io_u_plat;
1367
1368         int stride = 1 << hist_coarseness;
1369
1370         if (!sample_size)
1371                 return;
1372
1373         s = __get_sample(samples, 0, 0);
1374         log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
1375
1376         nr_samples = sample_size / __log_entry_sz(log_offset);
1377
1378         for (i = 0; i < nr_samples; i++) {
1379
1380                 s = (struct io_sample *)((char *)__get_sample(samples, log_offset, i) +
1381                         i * sizeof(struct io_u_plat_entry));
1382
1383                 entry = s->data.plat_entry;
1384                 io_u_plat = entry->io_u_plat;
1385
1386                 fprintf(f, "%lu, %u, %llu, ", (unsigned long) s->time,
1387                                                 io_sample_ddir(s), (unsigned long long) s->bs);
1388                 for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) {
1389                         fprintf(f, "%llu, ", (unsigned long long)hist_sum(j, stride, io_u_plat, NULL));
1390                 }
1391                 fprintf(f, "%llu\n", (unsigned long long)
1392                         hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat, NULL));
1393
1394         }
1395 }
1396
1397 static int fio_client_handle_iolog(struct fio_client *client,
1398                                    struct fio_net_cmd *cmd)
1399 {
1400         struct cmd_iolog_pdu *pdu = NULL;
1401         bool store_direct;
1402         char *log_pathname = NULL;
1403         int ret = 0;
1404
1405         pdu = convert_iolog(cmd, &store_direct);
1406         if (!pdu) {
1407                 log_err("fio: failed converting IO log\n");
1408                 ret = 1;
1409                 goto out;
1410         }
1411
1412         /* allocate buffer big enough for next sprintf() call */
1413         log_pathname = malloc(10 + strlen((char *)pdu->name) +
1414                         strlen(client->hostname));
1415         if (!log_pathname) {
1416                 log_err("fio: memory allocation of unique pathname failed\n");
1417                 ret = -1;
1418                 goto out;
1419         }
1420         /* generate a unique pathname for the log file using hostname */
1421         sprintf(log_pathname, "%s.%s", pdu->name, client->hostname);
1422
1423         if (store_direct) {
1424                 ssize_t wrote;
1425                 size_t sz;
1426                 int fd;
1427
1428                 fd = open((const char *) log_pathname,
1429                                 O_WRONLY | O_CREAT | O_TRUNC, 0644);
1430                 if (fd < 0) {
1431                         log_err("fio: open log %s: %s\n",
1432                                 log_pathname, strerror(errno));
1433                         ret = 1;
1434                         goto out;
1435                 }
1436
1437                 sz = cmd->pdu_len - sizeof(*pdu);
1438                 wrote = write(fd, pdu->samples, sz);
1439                 close(fd);
1440
1441                 if (wrote != sz) {
1442                         log_err("fio: short write on compressed log\n");
1443                         ret = 1;
1444                         goto out;
1445                 }
1446
1447                 ret = 0;
1448         } else {
1449                 FILE *f;
1450                 f = fopen((const char *) log_pathname, "w");
1451                 if (!f) {
1452                         log_err("fio: fopen log %s : %s\n",
1453                                 log_pathname, strerror(errno));
1454                         ret = 1;
1455                         goto out;
1456                 }
1457
1458                 if (pdu->log_type == IO_LOG_TYPE_HIST) {
1459                         client_flush_hist_samples(f, pdu->log_hist_coarseness, pdu->samples,
1460                                            pdu->nr_samples * sizeof(struct io_sample));
1461                 } else {
1462                         flush_samples(f, pdu->samples,
1463                                         pdu->nr_samples * sizeof(struct io_sample));
1464                 }
1465                 fclose(f);
1466                 ret = 0;
1467         }
1468
1469 out:
1470         if (pdu && pdu != (void *) cmd->payload)
1471                 free(pdu);
1472
1473         if (log_pathname)
1474                 free(log_pathname);
1475
1476         return ret;
1477 }
1478
1479 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
1480 {
1481         struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
1482         const char *os, *arch;
1483         char bit[16];
1484
1485         os = fio_get_os_string(probe->os);
1486         if (!os)
1487                 os = "unknown";
1488
1489         arch = fio_get_arch_string(probe->arch);
1490         if (!arch)
1491                 os = "unknown";
1492
1493         sprintf(bit, "%d-bit", probe->bpp * 8);
1494         probe->flags = le64_to_cpu(probe->flags);
1495
1496         if (output_format & FIO_OUTPUT_NORMAL) {
1497                 log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n",
1498                         probe->hostname, probe->bigendian, bit, os, arch,
1499                         probe->fio_version, (unsigned long) probe->flags);
1500         }
1501
1502         if (!client->name)
1503                 client->name = strdup((char *) probe->hostname);
1504 }
1505
1506 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
1507 {
1508         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1509
1510         client->state = Client_started;
1511         client->jobs = le32_to_cpu(pdu->jobs);
1512         client->nr_stat = le32_to_cpu(pdu->stat_outputs);
1513
1514         if (client->jobs) {
1515                 int i;
1516
1517                 if (client->opt_lists)
1518                         free(client->opt_lists);
1519
1520                 client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
1521                 for (i = 0; i < client->jobs; i++)
1522                         INIT_FLIST_HEAD(&client->opt_lists[i]);
1523         }
1524
1525         sum_stat_clients += client->nr_stat;
1526 }
1527
1528 static void handle_stop(struct fio_client *client)
1529 {
1530         if (client->error)
1531                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
1532 }
1533
1534 static void convert_stop(struct fio_net_cmd *cmd)
1535 {
1536         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1537
1538         pdu->error = le32_to_cpu(pdu->error);
1539 }
1540
1541 static void convert_text(struct fio_net_cmd *cmd)
1542 {
1543         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1544
1545         pdu->level      = le32_to_cpu(pdu->level);
1546         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
1547         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
1548         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
1549 }
1550
1551 static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
1552                                               struct cmd_iolog_pdu *pdu)
1553 {
1554 #ifdef CONFIG_ZLIB
1555         struct cmd_iolog_pdu *ret;
1556         z_stream stream;
1557         uint64_t nr_samples;
1558         size_t total;
1559         char *p;
1560
1561         stream.zalloc = Z_NULL;
1562         stream.zfree = Z_NULL;
1563         stream.opaque = Z_NULL;
1564         stream.avail_in = 0;
1565         stream.next_in = Z_NULL;
1566
1567         if (inflateInit(&stream) != Z_OK)
1568                 return NULL;
1569
1570         /*
1571          * Get header first, it's not compressed
1572          */
1573         nr_samples = le64_to_cpu(pdu->nr_samples);
1574
1575         if (pdu->log_type == IO_LOG_TYPE_HIST)
1576                 total = nr_samples * (__log_entry_sz(le32_to_cpu(pdu->log_offset)) +
1577                                         sizeof(struct io_u_plat_entry));
1578         else
1579                 total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
1580         ret = malloc(total + sizeof(*pdu));
1581         ret->nr_samples = nr_samples;
1582
1583         memcpy(ret, pdu, sizeof(*pdu));
1584
1585         p = (char *) ret + sizeof(*pdu);
1586
1587         stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1588         stream.next_in = (void *)((char *) pdu + sizeof(*pdu));
1589         while (stream.avail_in) {
1590                 unsigned int this_chunk = 65536;
1591                 unsigned int this_len;
1592                 int err;
1593
1594                 if (this_chunk > total)
1595                         this_chunk = total;
1596
1597                 stream.avail_out = this_chunk;
1598                 stream.next_out = (void *)p;
1599                 err = inflate(&stream, Z_NO_FLUSH);
1600                 /* may be Z_OK, or Z_STREAM_END */
1601                 if (err < 0) {
1602                         /*
1603                          * Z_STREAM_ERROR and Z_BUF_ERROR can safely be
1604                          * ignored */
1605                         if (err == Z_STREAM_ERROR || err == Z_BUF_ERROR)
1606                                 break;
1607                         log_err("fio: inflate error %d\n", err);
1608                         free(ret);
1609                         ret = NULL;
1610                         goto err;
1611                 }
1612
1613                 this_len = this_chunk - stream.avail_out;
1614                 p += this_len;
1615                 total -= this_len;
1616         }
1617
1618 err:
1619         inflateEnd(&stream);
1620         return ret;
1621 #else
1622         return NULL;
1623 #endif
1624 }
1625
1626 /*
1627  * This has been compressed on the server side, since it can be big.
1628  * Uncompress here.
1629  */
1630 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd,
1631                                            bool *store_direct)
1632 {
1633         struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1634         struct cmd_iolog_pdu *ret;
1635         uint64_t i;
1636         int compressed;
1637         void *samples;
1638
1639         *store_direct = false;
1640
1641         /*
1642          * Convert if compressed and we support it. If it's not
1643          * compressed, we need not do anything.
1644          */
1645         compressed = le32_to_cpu(pdu->compressed);
1646         if (compressed == XMIT_COMPRESSED) {
1647 #ifndef CONFIG_ZLIB
1648                 log_err("fio: server sent compressed data by mistake\n");
1649                 return NULL;
1650 #endif
1651                 ret = convert_iolog_gz(cmd, pdu);
1652                 if (!ret) {
1653                         log_err("fio: failed decompressing log\n");
1654                         return NULL;
1655                 }
1656         } else if (compressed == STORE_COMPRESSED) {
1657                 *store_direct = true;
1658                 ret = pdu;
1659         } else
1660                 ret = pdu;
1661
1662         ret->nr_samples         = le64_to_cpu(ret->nr_samples);
1663         ret->thread_number      = le32_to_cpu(ret->thread_number);
1664         ret->log_type           = le32_to_cpu(ret->log_type);
1665         ret->compressed         = le32_to_cpu(ret->compressed);
1666         ret->log_offset         = le32_to_cpu(ret->log_offset);
1667         ret->log_hist_coarseness = le32_to_cpu(ret->log_hist_coarseness);
1668
1669         if (*store_direct)
1670                 return ret;
1671
1672         samples = &ret->samples[0];
1673         for (i = 0; i < ret->nr_samples; i++) {
1674                 struct io_sample *s;
1675
1676                 s = __get_sample(samples, ret->log_offset, i);
1677                 if (ret->log_type == IO_LOG_TYPE_HIST)
1678                         s = (struct io_sample *)((char *)s + sizeof(struct io_u_plat_entry) * i);
1679
1680                 s->time         = le64_to_cpu(s->time);
1681                 s->data.val     = le64_to_cpu(s->data.val);
1682                 s->__ddir       = le32_to_cpu(s->__ddir);
1683                 s->bs           = le64_to_cpu(s->bs);
1684
1685                 if (ret->log_offset) {
1686                         struct io_sample_offset *so = (void *) s;
1687
1688                         so->offset = le64_to_cpu(so->offset);
1689                 }
1690
1691                 if (ret->log_type == IO_LOG_TYPE_HIST) {
1692                         s->data.plat_entry = (struct io_u_plat_entry *)(((char *)s) + sizeof(*s));
1693                         s->data.plat_entry->list.next = NULL;
1694                         s->data.plat_entry->list.prev = NULL;
1695                 }
1696         }
1697
1698         return ret;
1699 }
1700
1701 static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep,
1702                            size_t size, uint64_t tag)
1703 {
1704         rep->error = cpu_to_le32(rep->error);
1705         fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL);
1706 }
1707
1708 static int fio_send_file(struct fio_client *client, struct cmd_sendfile *pdu,
1709                          uint64_t tag)
1710 {
1711         struct cmd_sendfile_reply *rep;
1712         struct stat sb;
1713         size_t size;
1714         int fd;
1715
1716         size = sizeof(*rep);
1717         rep = malloc(size);
1718
1719         if (stat((char *)pdu->path, &sb) < 0) {
1720 fail:
1721                 rep->error = errno;
1722                 sendfile_reply(client->fd, rep, size, tag);
1723                 free(rep);
1724                 return 1;
1725         }
1726
1727         size += sb.st_size;
1728         rep = realloc(rep, size);
1729         rep->size = cpu_to_le32((uint32_t) sb.st_size);
1730
1731         fd = open((char *)pdu->path, O_RDONLY);
1732         if (fd == -1 )
1733                 goto fail;
1734
1735         rep->error = read_data(fd, &rep->data, sb.st_size);
1736         sendfile_reply(client->fd, rep, size, tag);
1737         free(rep);
1738         close(fd);
1739         return 0;
1740 }
1741
1742 int fio_handle_client(struct fio_client *client)
1743 {
1744         struct client_ops *ops = client->ops;
1745         struct fio_net_cmd *cmd;
1746         int size;
1747
1748         dprint(FD_NET, "client: handle %s\n", client->hostname);
1749
1750         cmd = fio_net_recv_cmd(client->fd, true);
1751         if (!cmd)
1752                 return 0;
1753
1754         dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1755                 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1756
1757         client->last_cmd = cmd->opcode;
1758
1759         switch (cmd->opcode) {
1760         case FIO_NET_CMD_QUIT:
1761                 if (ops->quit)
1762                         ops->quit(client, cmd);
1763                 remove_client(client);
1764                 break;
1765         case FIO_NET_CMD_TEXT:
1766                 convert_text(cmd);
1767                 ops->text(client, cmd);
1768                 break;
1769         case FIO_NET_CMD_DU: {
1770                 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1771
1772                 convert_dus(&du->dus);
1773                 convert_agg(&du->agg);
1774
1775                 ops->disk_util(client, cmd);
1776                 break;
1777                 }
1778         case FIO_NET_CMD_TS: {
1779                 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1780
1781                 dprint(FD_NET, "client: ts->ss_state = %u\n", (unsigned int) le32_to_cpu(p->ts.ss_state));
1782                 if (le32_to_cpu(p->ts.ss_state) & FIO_SS_DATA) {
1783                         dprint(FD_NET, "client: received steadystate ring buffers\n");
1784
1785                         size = le64_to_cpu(p->ts.ss_dur);
1786                         p->ts.ss_iops_data = (uint64_t *) ((struct cmd_ts_pdu *)cmd->payload + 1);
1787                         p->ts.ss_bw_data = p->ts.ss_iops_data + size;
1788                 }
1789
1790                 convert_ts(&p->ts, &p->ts);
1791                 convert_gs(&p->rs, &p->rs);
1792
1793                 ops->thread_status(client, cmd);
1794                 break;
1795                 }
1796         case FIO_NET_CMD_GS: {
1797                 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1798
1799                 convert_gs(gs, gs);
1800
1801                 ops->group_stats(client, cmd);
1802                 break;
1803                 }
1804         case FIO_NET_CMD_ETA: {
1805                 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1806
1807                 if (!remove_reply_cmd(client, cmd))
1808                         break;
1809                 convert_jobs_eta(je);
1810                 handle_eta(client, cmd);
1811                 break;
1812                 }
1813         case FIO_NET_CMD_PROBE:
1814                 remove_reply_cmd(client, cmd);
1815                 ops->probe(client, cmd);
1816                 break;
1817         case FIO_NET_CMD_SERVER_START:
1818                 client->state = Client_running;
1819                 if (ops->job_start)
1820                         ops->job_start(client, cmd);
1821                 break;
1822         case FIO_NET_CMD_START: {
1823                 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1824
1825                 pdu->jobs = le32_to_cpu(pdu->jobs);
1826                 ops->start(client, cmd);
1827                 break;
1828                 }
1829         case FIO_NET_CMD_STOP: {
1830                 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1831
1832                 convert_stop(cmd);
1833                 client->state = Client_stopped;
1834                 client->error = le32_to_cpu(pdu->error);
1835                 client->signal = le32_to_cpu(pdu->signal);
1836                 ops->stop(client);
1837                 break;
1838                 }
1839         case FIO_NET_CMD_ADD_JOB: {
1840                 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1841
1842                 client->thread_number = le32_to_cpu(pdu->thread_number);
1843                 client->groupid = le32_to_cpu(pdu->groupid);
1844
1845                 if (ops->add_job)
1846                         ops->add_job(client, cmd);
1847                 break;
1848                 }
1849         case FIO_NET_CMD_IOLOG:
1850                 fio_client_handle_iolog(client, cmd);
1851                 break;
1852         case FIO_NET_CMD_UPDATE_JOB:
1853                 ops->update_job(client, cmd);
1854                 remove_reply_cmd(client, cmd);
1855                 break;
1856         case FIO_NET_CMD_VTRIGGER: {
1857                 struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
1858                 char buf[128];
1859                 int off = 0;
1860
1861                 if (aux_path) {
1862                         strcpy(buf, aux_path);
1863                         off = strlen(buf);
1864                 }
1865
1866                 __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
1867                 exec_trigger(trigger_cmd);
1868                 break;
1869                 }
1870         case FIO_NET_CMD_SENDFILE: {
1871                 struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload;
1872                 fio_send_file(client, pdu, cmd->tag);
1873                 break;
1874                 }
1875         case FIO_NET_CMD_JOB_OPT: {
1876                 handle_job_opt(client, cmd);
1877                 break;
1878         }
1879         default:
1880                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1881                 break;
1882         }
1883
1884         free(cmd);
1885         return 1;
1886 }
1887
1888 int fio_clients_send_trigger(const char *cmd)
1889 {
1890         struct flist_head *entry;
1891         struct fio_client *client;
1892         size_t slen;
1893
1894         dprint(FD_NET, "client: send vtrigger: %s\n", cmd);
1895
1896         if (!cmd)
1897                 slen = 0;
1898         else
1899                 slen = strlen(cmd);
1900
1901         flist_for_each(entry, &client_list) {
1902                 struct cmd_vtrigger_pdu *pdu;
1903
1904                 client = flist_entry(entry, struct fio_client, list);
1905
1906                 pdu = malloc(sizeof(*pdu) + slen);
1907                 pdu->len = cpu_to_le16((uint16_t) slen);
1908                 if (slen)
1909                         memcpy(pdu->cmd, cmd, slen);
1910                 fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu,
1911                                         sizeof(*pdu) + slen, NULL, NULL);
1912                 free(pdu);
1913         }
1914
1915         return 0;
1916 }
1917
1918 static void request_client_etas(struct client_ops *ops)
1919 {
1920         struct fio_client *client;
1921         struct flist_head *entry;
1922         struct client_eta *eta;
1923         int skipped = 0;
1924
1925         if (eta_print == FIO_ETA_NEVER)
1926                 return;
1927
1928         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1929
1930         eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
1931         eta->pending = nr_clients;
1932
1933         flist_for_each(entry, &client_list) {
1934                 client = flist_entry(entry, struct fio_client, list);
1935
1936                 if (!flist_empty(&client->eta_list)) {
1937                         skipped++;
1938                         continue;
1939                 }
1940                 if (client->state != Client_running)
1941                         continue;
1942
1943                 assert(!client->eta_in_flight);
1944                 flist_add_tail(&client->eta_list, &eta_list);
1945                 client->eta_in_flight = eta;
1946                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1947                                         (uintptr_t) eta, &client->cmd_list);
1948         }
1949
1950         while (skipped--) {
1951                 if (!fio_client_dec_jobs_eta(eta, ops->eta))
1952                         break;
1953         }
1954
1955         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1956 }
1957
1958 /*
1959  * A single SEND_ETA timeout isn't fatal. Attempt to recover.
1960  */
1961 static int handle_cmd_timeout(struct fio_client *client,
1962                               struct fio_net_cmd_reply *reply)
1963 {
1964         uint16_t reply_opcode = reply->opcode;
1965
1966         flist_del(&reply->list);
1967         free(reply);
1968
1969         if (reply_opcode != FIO_NET_CMD_SEND_ETA)
1970                 return 1;
1971
1972         log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
1973
1974         flist_del_init(&client->eta_list);
1975         if (client->eta_in_flight) {
1976                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
1977                 client->eta_in_flight = NULL;
1978         }
1979
1980         /*
1981          * If we fail 5 in a row, give up...
1982          */
1983         if (client->eta_timeouts++ > 5)
1984                 return 1;
1985
1986         return 0;
1987 }
1988
1989 static int client_check_cmd_timeout(struct fio_client *client,
1990                                     struct timespec *now)
1991 {
1992         struct fio_net_cmd_reply *reply;
1993         struct flist_head *entry, *tmp;
1994         int ret = 0;
1995
1996         flist_for_each_safe(entry, tmp, &client->cmd_list) {
1997                 unsigned int op;
1998
1999                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
2000
2001                 if (mtime_since(&reply->ts, now) < FIO_NET_CLIENT_TIMEOUT)
2002                         continue;
2003
2004                 op = reply->opcode;
2005                 if (!handle_cmd_timeout(client, reply))
2006                         continue;
2007
2008                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
2009                                                 fio_server_op(op));
2010                 ret = 1;
2011         }
2012
2013         return flist_empty(&client->cmd_list) && ret;
2014 }
2015
2016 static int fio_check_clients_timed_out(void)
2017 {
2018         struct fio_client *client;
2019         struct flist_head *entry, *tmp;
2020         struct timespec ts;
2021         int ret = 0;
2022
2023         fio_gettime(&ts, NULL);
2024
2025         flist_for_each_safe(entry, tmp, &client_list) {
2026                 client = flist_entry(entry, struct fio_client, list);
2027
2028                 if (flist_empty(&client->cmd_list))
2029                         continue;
2030
2031                 if (!client_check_cmd_timeout(client, &ts))
2032                         continue;
2033
2034                 if (client->ops->timed_out)
2035                         client->ops->timed_out(client);
2036                 else
2037                         log_err("fio: client %s timed out\n", client->hostname);
2038
2039                 if (client->last_cmd != FIO_NET_CMD_VTRIGGER)
2040                         client->error = ETIMEDOUT;
2041                 else
2042                         log_info("fio: ignoring timeout due to vtrigger\n");
2043                 remove_client(client);
2044                 ret = 1;
2045         }
2046
2047         return ret;
2048 }
2049
2050 int fio_handle_clients(struct client_ops *ops)
2051 {
2052         struct pollfd *pfds;
2053         int i, ret = 0, retval = 0;
2054
2055         fio_gettime(&eta_ts, NULL);
2056
2057         pfds = malloc(nr_clients * sizeof(struct pollfd));
2058
2059         init_thread_stat(&client_ts);
2060         init_group_run_stat(&client_gs);
2061
2062         while (!exit_backend && nr_clients) {
2063                 struct flist_head *entry, *tmp;
2064                 struct fio_client *client;
2065
2066                 i = 0;
2067                 flist_for_each_safe(entry, tmp, &client_list) {
2068                         client = flist_entry(entry, struct fio_client, list);
2069
2070                         if (!client->sent_job && !client->ops->stay_connected &&
2071                             flist_empty(&client->cmd_list)) {
2072                                 remove_client(client);
2073                                 continue;
2074                         }
2075
2076                         pfds[i].fd = client->fd;
2077                         pfds[i].events = POLLIN;
2078                         i++;
2079                 }
2080
2081                 if (!nr_clients)
2082                         break;
2083
2084                 assert(i == nr_clients);
2085
2086                 do {
2087                         struct timespec ts;
2088                         int timeout;
2089
2090                         fio_gettime(&ts, NULL);
2091                         if (eta_time_within_slack(mtime_since(&eta_ts, &ts))) {
2092                                 request_client_etas(ops);
2093                                 memcpy(&eta_ts, &ts, sizeof(ts));
2094
2095                                 if (fio_check_clients_timed_out())
2096                                         break;
2097                         }
2098
2099                         check_trigger_file();
2100
2101                         timeout = min(100u, ops->eta_msec);
2102
2103                         ret = poll(pfds, nr_clients, timeout);
2104                         if (ret < 0) {
2105                                 if (errno == EINTR)
2106                                         continue;
2107                                 log_err("fio: poll clients: %s\n", strerror(errno));
2108                                 break;
2109                         } else if (!ret)
2110                                 continue;
2111                 } while (ret <= 0);
2112
2113                 for (i = 0; i < nr_clients; i++) {
2114                         if (!(pfds[i].revents & POLLIN))
2115                                 continue;
2116
2117                         client = find_client_by_fd(pfds[i].fd);
2118                         if (!client) {
2119                                 log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd);
2120                                 continue;
2121                         }
2122                         if (!fio_handle_client(client)) {
2123                                 log_info("client: host=%s disconnected\n",
2124                                                 client->hostname);
2125                                 remove_client(client);
2126                                 retval = 1;
2127                         } else if (client->error)
2128                                 retval = 1;
2129                         fio_put_client(client);
2130                 }
2131         }
2132
2133         log_info_buf(allclients.buf, allclients.buflen);
2134         buf_output_free(&allclients);
2135
2136         fio_client_json_fini();
2137
2138         free(pfds);
2139         return retval || error_clients;
2140 }
2141
2142 static void client_display_thread_status(struct jobs_eta *je)
2143 {
2144         if (!(output_format & FIO_OUTPUT_JSON))
2145                 display_thread_status(je);
2146 }