Merge branch 'master' of https://github.com/bvanassche/fio into master
[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         if (type == Fio_client_socket)
394                 client->is_sock = true;
395         else {
396                 int ipv6;
397
398                 ipv6 = type == Fio_client_ipv6;
399                 if (fio_server_parse_host(hostname, ipv6,
400                                                 &client->addr.sin_addr,
401                                                 &client->addr6.sin6_addr))
402                         goto err;
403
404                 client->port = port;
405         }
406
407         client->fd = -1;
408         client->ops = ops;
409         client->refs = 1;
410         client->type = ops->client_type;
411         client->hostname = strdup(hostname);
412
413         __fio_client_add_cmd_option(client, "fio");
414
415         flist_add(&client->list, &client_list);
416         nr_clients++;
417         dprint(FD_NET, "client: added <%s>\n", client->hostname);
418         return client;
419 err:
420         free(client);
421         return NULL;
422 }
423
424 int fio_client_add_ini_file(void *cookie, const char *ini_file, bool remote)
425 {
426         struct fio_client *client = cookie;
427         struct client_file *cf;
428         size_t new_size;
429         void *new_files;
430
431         if (!client)
432                 return 1;
433
434         dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file);
435
436         new_size = (client->nr_files + 1) * sizeof(struct client_file);
437         new_files = realloc(client->files, new_size);
438         if (!new_files)
439                 return 1;
440
441         client->files = new_files;
442         cf = &client->files[client->nr_files];
443         cf->file = strdup(ini_file);
444         cf->remote = remote;
445         client->nr_files++;
446         return 0;
447 }
448
449 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
450 {
451         struct fio_client *existing = *cookie;
452         struct fio_client *client;
453
454         if (existing) {
455                 /*
456                  * We always add our "exec" name as the option, hence 1
457                  * means empty.
458                  */
459                 if (existing->argc == 1)
460                         flist_add_tail(&existing->arg_list, &arg_list);
461                 else {
462                         while (!flist_empty(&arg_list))
463                                 flist_del_init(arg_list.next);
464                 }
465         }
466
467         client = get_new_client();
468
469         if (fio_server_parse_string(hostname, &client->hostname,
470                                         &client->is_sock, &client->port,
471                                         &client->addr.sin_addr,
472                                         &client->addr6.sin6_addr,
473                                         &client->ipv6)) {
474                 fio_put_client(client);
475                 return -1;
476         }
477
478         client->fd = -1;
479         client->ops = ops;
480         client->refs = 1;
481         client->type = ops->client_type;
482
483         __fio_client_add_cmd_option(client, "fio");
484
485         flist_add(&client->list, &client_list);
486         nr_clients++;
487         dprint(FD_NET, "client: added <%s>\n", client->hostname);
488         *cookie = client;
489         return 0;
490 }
491
492 static const char *server_name(struct fio_client *client, char *buf,
493                                size_t bufsize)
494 {
495         const char *from;
496
497         if (client->ipv6)
498                 from = inet_ntop(AF_INET6, (struct sockaddr *) &client->addr6.sin6_addr, buf, bufsize);
499         else if (client->is_sock)
500                 from = "sock";
501         else
502                 from = inet_ntop(AF_INET, (struct sockaddr *) &client->addr.sin_addr, buf, bufsize);
503
504         return from;
505 }
506
507 static void probe_client(struct fio_client *client)
508 {
509         struct cmd_client_probe_pdu pdu;
510         const char *sname;
511         uint64_t tag;
512         char buf[64];
513
514         dprint(FD_NET, "client: send probe\n");
515
516 #ifdef CONFIG_ZLIB
517         pdu.flags = __le64_to_cpu(FIO_PROBE_FLAG_ZLIB);
518 #else
519         pdu.flags = 0;
520 #endif
521
522         sname = server_name(client, buf, sizeof(buf));
523         memset(pdu.server, 0, sizeof(pdu.server));
524         snprintf((char *) pdu.server, sizeof(pdu.server), "%s", sname);
525
526         fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list);
527 }
528
529 static int fio_client_connect_ip(struct fio_client *client)
530 {
531         struct sockaddr *addr;
532         socklen_t socklen;
533         int fd, domain;
534
535         if (client->ipv6) {
536                 client->addr6.sin6_family = AF_INET6;
537                 client->addr6.sin6_port = htons(client->port);
538                 domain = AF_INET6;
539                 addr = (struct sockaddr *) &client->addr6;
540                 socklen = sizeof(client->addr6);
541         } else {
542                 client->addr.sin_family = AF_INET;
543                 client->addr.sin_port = htons(client->port);
544                 domain = AF_INET;
545                 addr = (struct sockaddr *) &client->addr;
546                 socklen = sizeof(client->addr);
547         }
548
549         fd = socket(domain, SOCK_STREAM, 0);
550         if (fd < 0) {
551                 int ret = -errno;
552
553                 log_err("fio: socket: %s\n", strerror(errno));
554                 return ret;
555         }
556
557         if (connect(fd, addr, socklen) < 0) {
558                 int ret = -errno;
559
560                 log_err("fio: connect: %s\n", strerror(errno));
561                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
562                                                                 client->port);
563                 close(fd);
564                 return ret;
565         }
566
567         return fd;
568 }
569
570 static int fio_client_connect_sock(struct fio_client *client)
571 {
572         struct sockaddr_un *addr = &client->addr_un;
573         socklen_t len;
574         int fd;
575
576         memset(addr, 0, sizeof(*addr));
577         addr->sun_family = AF_UNIX;
578         snprintf(addr->sun_path, sizeof(addr->sun_path), "%s",
579                  client->hostname);
580
581         fd = socket(AF_UNIX, SOCK_STREAM, 0);
582         if (fd < 0) {
583                 int ret = -errno;
584
585                 log_err("fio: socket: %s\n", strerror(errno));
586                 return ret;
587         }
588
589         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
590         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
591                 int ret = -errno;
592
593                 log_err("fio: connect; %s\n", strerror(errno));
594                 close(fd);
595                 return ret;
596         }
597
598         return fd;
599 }
600
601 int fio_client_connect(struct fio_client *client)
602 {
603         int fd;
604
605         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
606
607         if (client->is_sock)
608                 fd = fio_client_connect_sock(client);
609         else
610                 fd = fio_client_connect_ip(client);
611
612         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
613
614         if (fd < 0)
615                 return fd;
616
617         client->fd = fd;
618         fio_client_add_hash(client);
619         client->state = Client_connected;
620
621         probe_client(client);
622         return 0;
623 }
624
625 int fio_client_terminate(struct fio_client *client)
626 {
627         return fio_net_send_quit(client->fd);
628 }
629
630 static void fio_clients_terminate(void)
631 {
632         struct flist_head *entry;
633         struct fio_client *client;
634
635         dprint(FD_NET, "client: terminate clients\n");
636
637         flist_for_each(entry, &client_list) {
638                 client = flist_entry(entry, struct fio_client, list);
639                 fio_client_terminate(client);
640         }
641 }
642
643 static void sig_int(int sig)
644 {
645         dprint(FD_NET, "client: got signal %d\n", sig);
646         fio_clients_terminate();
647 }
648
649 static void client_signal_handler(void)
650 {
651         struct sigaction act;
652
653         memset(&act, 0, sizeof(act));
654         act.sa_handler = sig_int;
655         act.sa_flags = SA_RESTART;
656         sigaction(SIGINT, &act, NULL);
657
658         memset(&act, 0, sizeof(act));
659         act.sa_handler = sig_int;
660         act.sa_flags = SA_RESTART;
661         sigaction(SIGTERM, &act, NULL);
662
663 /* Windows uses SIGBREAK as a quit signal from other applications */
664 #ifdef WIN32
665         memset(&act, 0, sizeof(act));
666         act.sa_handler = sig_int;
667         act.sa_flags = SA_RESTART;
668         sigaction(SIGBREAK, &act, NULL);
669 #endif
670
671         memset(&act, 0, sizeof(act));
672         act.sa_handler = sig_show_status;
673         act.sa_flags = SA_RESTART;
674         sigaction(SIGUSR1, &act, NULL);
675 }
676
677 static int send_client_cmd_line(struct fio_client *client)
678 {
679         struct cmd_single_line_pdu *cslp;
680         struct cmd_line_pdu *clp;
681         unsigned long offset;
682         unsigned int *lens;
683         void *pdu;
684         size_t mem;
685         int i, ret;
686
687         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
688
689         lens = malloc(client->argc * sizeof(unsigned int));
690
691         /*
692          * Find out how much mem we need
693          */
694         for (i = 0, mem = 0; i < client->argc; i++) {
695                 lens[i] = strlen(client->argv[i]) + 1;
696                 mem += lens[i];
697         }
698
699         /*
700          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
701          */
702         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
703
704         pdu = malloc(mem);
705         clp = pdu;
706         offset = sizeof(*clp);
707
708         for (i = 0; i < client->argc; i++) {
709                 uint16_t arg_len = lens[i];
710
711                 cslp = pdu + offset;
712                 strcpy((char *) cslp->text, client->argv[i]);
713                 cslp->len = cpu_to_le16(arg_len);
714                 offset += sizeof(*cslp) + arg_len;
715         }
716
717         free(lens);
718         clp->lines = cpu_to_le16(client->argc);
719         clp->client_type = __cpu_to_le16(client->type);
720         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, NULL, NULL);
721         free(pdu);
722         return ret;
723 }
724
725 int fio_clients_connect(void)
726 {
727         struct fio_client *client;
728         struct flist_head *entry, *tmp;
729         int ret;
730
731 #ifdef WIN32
732         WSADATA wsd;
733         WSAStartup(MAKEWORD(2, 2), &wsd);
734 #endif
735
736         dprint(FD_NET, "client: connect all\n");
737
738         client_signal_handler();
739
740         flist_for_each_safe(entry, tmp, &client_list) {
741                 client = flist_entry(entry, struct fio_client, list);
742
743                 ret = fio_client_connect(client);
744                 if (ret) {
745                         remove_client(client);
746                         continue;
747                 }
748
749                 if (client->argc > 1)
750                         send_client_cmd_line(client);
751         }
752
753         return !nr_clients;
754 }
755
756 int fio_start_client(struct fio_client *client)
757 {
758         dprint(FD_NET, "client: start %s\n", client->hostname);
759         return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
760 }
761
762 int fio_start_all_clients(void)
763 {
764         struct fio_client *client;
765         struct flist_head *entry, *tmp;
766         int ret;
767
768         dprint(FD_NET, "client: start all\n");
769
770         fio_client_json_init();
771
772         flist_for_each_safe(entry, tmp, &client_list) {
773                 client = flist_entry(entry, struct fio_client, list);
774
775                 ret = fio_start_client(client);
776                 if (ret) {
777                         remove_client(client);
778                         continue;
779                 }
780         }
781
782         return flist_empty(&client_list);
783 }
784
785 static int __fio_client_send_remote_ini(struct fio_client *client,
786                                         const char *filename)
787 {
788         struct cmd_load_file_pdu *pdu;
789         size_t p_size;
790         int ret;
791
792         dprint(FD_NET, "send remote ini %s to %s\n", filename, client->hostname);
793
794         p_size = sizeof(*pdu) + strlen(filename) + 1;
795         pdu = malloc(p_size);
796         memset(pdu, 0, p_size);
797         pdu->name_len = strlen(filename);
798         strcpy((char *) pdu->file, filename);
799         pdu->client_type = cpu_to_le16((uint16_t) client->type);
800
801         client->sent_job = true;
802         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_LOAD_FILE, pdu, p_size,NULL, NULL);
803         free(pdu);
804         return ret;
805 }
806
807 /*
808  * Send file contents to server backend. We could use sendfile(), but to remain
809  * more portable lets just read/write the darn thing.
810  */
811 static int __fio_client_send_local_ini(struct fio_client *client,
812                                        const char *filename)
813 {
814         struct cmd_job_pdu *pdu;
815         size_t p_size;
816         struct stat sb;
817         char *p;
818         void *buf;
819         off_t len;
820         int fd, ret;
821
822         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
823
824         fd = open(filename, O_RDONLY);
825         if (fd < 0) {
826                 ret = -errno;
827                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
828                 return ret;
829         }
830
831         if (fstat(fd, &sb) < 0) {
832                 ret = -errno;
833                 log_err("fio: job file stat: %s\n", strerror(errno));
834                 close(fd);
835                 return ret;
836         }
837
838         /*
839          * Add extra space for variable expansion, but doesn't guarantee.
840          */
841         sb.st_size += OPT_LEN_MAX;
842         p_size = sb.st_size + sizeof(*pdu);
843         pdu = malloc(p_size);
844         buf = pdu->buf;
845
846         len = sb.st_size;
847         p = buf;
848         if (read_ini_data(fd, p, len)) {
849                 log_err("fio: failed reading job file %s\n", filename);
850                 close(fd);
851                 free(pdu);
852                 return 1;
853         }
854
855         pdu->buf_len = __cpu_to_le32(sb.st_size);
856         pdu->client_type = cpu_to_le32(client->type);
857
858         client->sent_job = true;
859         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, NULL, NULL);
860         free(pdu);
861         close(fd);
862         return ret;
863 }
864
865 int fio_client_send_ini(struct fio_client *client, const char *filename,
866                         bool remote)
867 {
868         int ret;
869
870         if (!remote)
871                 ret = __fio_client_send_local_ini(client, filename);
872         else
873                 ret = __fio_client_send_remote_ini(client, filename);
874
875         if (!ret)
876                 client->sent_job = true;
877
878         return ret;
879 }
880
881 static int fio_client_send_cf(struct fio_client *client,
882                               struct client_file *cf)
883 {
884         return fio_client_send_ini(client, cf->file, cf->remote);
885 }
886
887 int fio_clients_send_ini(const char *filename)
888 {
889         struct fio_client *client;
890         struct flist_head *entry, *tmp;
891
892         flist_for_each_safe(entry, tmp, &client_list) {
893                 bool failed = false;
894
895                 client = flist_entry(entry, struct fio_client, list);
896
897                 if (client->nr_files) {
898                         int i;
899
900                         for (i = 0; i < client->nr_files; i++) {
901                                 struct client_file *cf;
902
903                                 cf = &client->files[i];
904
905                                 if (fio_client_send_cf(client, cf)) {
906                                         failed = true;
907                                         remove_client(client);
908                                         break;
909                                 }
910                         }
911                 }
912                 if (client->sent_job || failed)
913                         continue;
914                 if (!filename || fio_client_send_ini(client, filename, 0))
915                         remove_client(client);
916         }
917
918         return !nr_clients;
919 }
920
921 int fio_client_update_options(struct fio_client *client,
922                               struct thread_options *o, uint64_t *tag)
923 {
924         struct cmd_add_job_pdu pdu;
925
926         pdu.thread_number = cpu_to_le32(client->thread_number);
927         pdu.groupid = cpu_to_le32(client->groupid);
928         convert_thread_options_to_net(&pdu.top, o);
929
930         return fio_net_send_cmd(client->fd, FIO_NET_CMD_UPDATE_JOB, &pdu, sizeof(pdu), tag, &client->cmd_list);
931 }
932
933 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
934 {
935         dst->max_val    = le64_to_cpu(src->max_val);
936         dst->min_val    = le64_to_cpu(src->min_val);
937         dst->samples    = le64_to_cpu(src->samples);
938
939         /*
940          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
941          */
942         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
943         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
944 }
945
946 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
947 {
948         int i, j, k;
949
950         dst->error              = le32_to_cpu(src->error);
951         dst->thread_number      = le32_to_cpu(src->thread_number);
952         dst->groupid            = le32_to_cpu(src->groupid);
953         dst->pid                = le32_to_cpu(src->pid);
954         dst->members            = le32_to_cpu(src->members);
955         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
956
957         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
958                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
959                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
960                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
961                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
962                 convert_io_stat(&dst->iops_stat[i], &src->iops_stat[i]);
963         }
964         convert_io_stat(&dst->sync_stat, &src->sync_stat);
965
966         dst->usr_time           = le64_to_cpu(src->usr_time);
967         dst->sys_time           = le64_to_cpu(src->sys_time);
968         dst->ctx                = le64_to_cpu(src->ctx);
969         dst->minf               = le64_to_cpu(src->minf);
970         dst->majf               = le64_to_cpu(src->majf);
971         dst->clat_percentiles   = le32_to_cpu(src->clat_percentiles);
972         dst->lat_percentiles    = le32_to_cpu(src->lat_percentiles);
973         dst->slat_percentiles   = le32_to_cpu(src->slat_percentiles);
974         dst->percentile_precision = le64_to_cpu(src->percentile_precision);
975
976         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
977                 fio_fp64_t *fps = &src->percentile_list[i];
978                 fio_fp64_t *fpd = &dst->percentile_list[i];
979
980                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
981         }
982
983         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
984                 dst->io_u_map[i]        = le64_to_cpu(src->io_u_map[i]);
985                 dst->io_u_submit[i]     = le64_to_cpu(src->io_u_submit[i]);
986                 dst->io_u_complete[i]   = le64_to_cpu(src->io_u_complete[i]);
987         }
988
989         for (i = 0; i < FIO_IO_U_LAT_N_NR; i++)
990                 dst->io_u_lat_n[i]      = le64_to_cpu(src->io_u_lat_n[i]);
991         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
992                 dst->io_u_lat_u[i]      = le64_to_cpu(src->io_u_lat_u[i]);
993         for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
994                 dst->io_u_lat_m[i]      = le64_to_cpu(src->io_u_lat_m[i]);
995
996         for (i = 0; i < FIO_LAT_CNT; i++)
997                 for (j = 0; j < DDIR_RWDIR_CNT; j++)
998                         for (k = 0; k < FIO_IO_U_PLAT_NR; k++)
999                                 dst->io_u_plat[i][j][k] = le64_to_cpu(src->io_u_plat[i][j][k]);
1000
1001         for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
1002                 dst->io_u_sync_plat[j] = le64_to_cpu(src->io_u_sync_plat[j]);
1003
1004         for (i = 0; i < DDIR_RWDIR_SYNC_CNT; i++)
1005                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
1006
1007         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1008                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
1009                 dst->drop_io_u[i]       = le64_to_cpu(src->drop_io_u[i]);
1010         }
1011
1012         dst->total_submit       = le64_to_cpu(src->total_submit);
1013         dst->total_complete     = le64_to_cpu(src->total_complete);
1014         dst->nr_zone_resets     = le64_to_cpu(src->nr_zone_resets);
1015
1016         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1017                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
1018                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
1019         }
1020
1021         dst->total_run_time     = le64_to_cpu(src->total_run_time);
1022         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
1023         dst->total_err_count    = le64_to_cpu(src->total_err_count);
1024         dst->first_error        = le32_to_cpu(src->first_error);
1025         dst->kb_base            = le32_to_cpu(src->kb_base);
1026         dst->unit_base          = le32_to_cpu(src->unit_base);
1027
1028         dst->sig_figs           = le32_to_cpu(src->sig_figs);
1029
1030         dst->latency_depth      = le32_to_cpu(src->latency_depth);
1031         dst->latency_target     = le64_to_cpu(src->latency_target);
1032         dst->latency_window     = le64_to_cpu(src->latency_window);
1033         dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
1034
1035         dst->nr_block_infos     = le64_to_cpu(src->nr_block_infos);
1036         for (i = 0; i < dst->nr_block_infos; i++)
1037                 dst->block_infos[i] = le32_to_cpu(src->block_infos[i]);
1038         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1039                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++) {
1040                         dst->io_u_plat_high_prio[i][j] = le64_to_cpu(src->io_u_plat_high_prio[i][j]);
1041                         dst->io_u_plat_low_prio[i][j] = le64_to_cpu(src->io_u_plat_low_prio[i][j]);
1042                 }
1043                 convert_io_stat(&dst->clat_high_prio_stat[i], &src->clat_high_prio_stat[i]);
1044                 convert_io_stat(&dst->clat_low_prio_stat[i], &src->clat_low_prio_stat[i]);
1045         }
1046
1047         dst->ss_dur             = le64_to_cpu(src->ss_dur);
1048         dst->ss_state           = le32_to_cpu(src->ss_state);
1049         dst->ss_head            = le32_to_cpu(src->ss_head);
1050         dst->ss_limit.u.f       = fio_uint64_to_double(le64_to_cpu(src->ss_limit.u.i));
1051         dst->ss_slope.u.f       = fio_uint64_to_double(le64_to_cpu(src->ss_slope.u.i));
1052         dst->ss_deviation.u.f   = fio_uint64_to_double(le64_to_cpu(src->ss_deviation.u.i));
1053         dst->ss_criterion.u.f   = fio_uint64_to_double(le64_to_cpu(src->ss_criterion.u.i));
1054
1055         if (dst->ss_state & FIO_SS_DATA) {
1056                 for (i = 0; i < dst->ss_dur; i++ ) {
1057                         dst->ss_iops_data[i] = le64_to_cpu(src->ss_iops_data[i]);
1058                         dst->ss_bw_data[i] = le64_to_cpu(src->ss_bw_data[i]);
1059                 }
1060         }
1061
1062         dst->cachehit           = le64_to_cpu(src->cachehit);
1063         dst->cachemiss          = le64_to_cpu(src->cachemiss);
1064 }
1065
1066 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
1067 {
1068         int i;
1069
1070         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1071                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
1072                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
1073                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
1074                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
1075                 dst->iobytes[i]         = le64_to_cpu(src->iobytes[i]);
1076                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
1077         }
1078
1079         dst->kb_base    = le32_to_cpu(src->kb_base);
1080         dst->unit_base  = le32_to_cpu(src->unit_base);
1081         dst->sig_figs   = le32_to_cpu(src->sig_figs);
1082         dst->groupid    = le32_to_cpu(src->groupid);
1083         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
1084 }
1085
1086 static void json_object_add_client_info(struct json_object *obj,
1087                                         struct fio_client *client)
1088 {
1089         const char *hostname = client->hostname ? client->hostname : "";
1090
1091         json_object_add_value_string(obj, "hostname", hostname);
1092         json_object_add_value_int(obj, "port", client->port);
1093 }
1094
1095 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
1096 {
1097         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1098         struct flist_head *opt_list = NULL;
1099         struct json_object *tsobj;
1100
1101         if (client->opt_lists && p->ts.thread_number <= client->jobs)
1102                 opt_list = &client->opt_lists[p->ts.thread_number - 1];
1103
1104         tsobj = show_thread_status(&p->ts, &p->rs, opt_list, &client->buf);
1105         client->did_stat = true;
1106         if (tsobj) {
1107                 json_object_add_client_info(tsobj, client);
1108                 json_array_add_value_object(clients_array, tsobj);
1109         }
1110
1111         if (sum_stat_clients <= 1)
1112                 return;
1113
1114         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1);
1115         sum_group_stats(&client_gs, &p->rs);
1116
1117         client_ts.members++;
1118         client_ts.thread_number = p->ts.thread_number;
1119         client_ts.groupid = p->ts.groupid;
1120         client_ts.unified_rw_rep = p->ts.unified_rw_rep;
1121         client_ts.sig_figs = p->ts.sig_figs;
1122
1123         if (++sum_stat_nr == sum_stat_clients) {
1124                 strcpy(client_ts.name, "All clients");
1125                 tsobj = show_thread_status(&client_ts, &client_gs, NULL, &allclients);
1126                 if (tsobj) {
1127                         json_object_add_client_info(tsobj, client);
1128                         json_array_add_value_object(clients_array, tsobj);
1129                 }
1130         }
1131 }
1132
1133 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
1134 {
1135         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1136
1137         if (output_format & FIO_OUTPUT_NORMAL)
1138                 show_group_stats(gs, &client->buf);
1139 }
1140
1141 static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
1142 {
1143         struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
1144
1145         pdu->global = le16_to_cpu(pdu->global);
1146         pdu->truncated = le16_to_cpu(pdu->truncated);
1147         pdu->groupid = le32_to_cpu(pdu->groupid);
1148
1149         if (pdu->global) {
1150                 if (!job_opt_object)
1151                         return;
1152
1153                 json_object_add_value_string(job_opt_object,
1154                                              (const char *)pdu->name,
1155                                              (const char *)pdu->value);
1156         } else if (client->opt_lists) {
1157                 struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
1158                 struct print_option *p;
1159
1160                 p = malloc(sizeof(*p));
1161                 p->name = strdup((const char *)pdu->name);
1162                 p->value = pdu->value[0] ? strdup((const char *)pdu->value) :
1163                         NULL;
1164                 flist_add_tail(&p->list, opt_list);
1165         }
1166 }
1167
1168 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
1169 {
1170         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1171         const char *buf = (const char *) pdu->buf;
1172         const char *name;
1173         int fio_unused ret;
1174         struct buf_output out;
1175
1176         buf_output_init(&out);
1177
1178         name = client->name ? client->name : client->hostname;
1179
1180         if (!client->skip_newline && !(output_format & FIO_OUTPUT_TERSE))
1181                 __log_buf(&out, "<%s> ", name);
1182         __log_buf(&out, "%s", buf);
1183         log_info_buf(out.buf, out.buflen);
1184         buf_output_free(&out);
1185         client->skip_newline = strchr(buf, '\n') == NULL;
1186 }
1187
1188 static void convert_agg(struct disk_util_agg *agg)
1189 {
1190         int i;
1191
1192         for (i = 0; i < 2; i++) {
1193                 agg->ios[i]     = le64_to_cpu(agg->ios[i]);
1194                 agg->merges[i]  = le64_to_cpu(agg->merges[i]);
1195                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
1196                 agg->ticks[i]   = le64_to_cpu(agg->ticks[i]);
1197         }
1198
1199         agg->io_ticks           = le64_to_cpu(agg->io_ticks);
1200         agg->time_in_queue      = le64_to_cpu(agg->time_in_queue);
1201         agg->slavecount         = le32_to_cpu(agg->slavecount);
1202         agg->max_util.u.f       = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
1203 }
1204
1205 static void convert_dus(struct disk_util_stat *dus)
1206 {
1207         int i;
1208
1209         for (i = 0; i < 2; i++) {
1210                 dus->s.ios[i]           = le64_to_cpu(dus->s.ios[i]);
1211                 dus->s.merges[i]        = le64_to_cpu(dus->s.merges[i]);
1212                 dus->s.sectors[i]       = le64_to_cpu(dus->s.sectors[i]);
1213                 dus->s.ticks[i]         = le64_to_cpu(dus->s.ticks[i]);
1214         }
1215
1216         dus->s.io_ticks         = le64_to_cpu(dus->s.io_ticks);
1217         dus->s.time_in_queue    = le64_to_cpu(dus->s.time_in_queue);
1218         dus->s.msec             = le64_to_cpu(dus->s.msec);
1219 }
1220
1221 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
1222 {
1223         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1224
1225         if (!client->disk_stats_shown)
1226                 client->disk_stats_shown = true;
1227
1228         if (output_format & FIO_OUTPUT_JSON) {
1229                 struct json_object *duobj;
1230
1231                 json_array_add_disk_util(&du->dus, &du->agg, du_array);
1232                 duobj = json_array_last_value_object(du_array);
1233                 json_object_add_client_info(duobj, client);
1234         }
1235         if (output_format & FIO_OUTPUT_NORMAL) {
1236                 __log_buf(&client->buf, "\nDisk stats (read/write):\n");
1237                 print_disk_util(&du->dus, &du->agg, 0, &client->buf);
1238         }
1239         if (output_format & FIO_OUTPUT_TERSE && terse_version >= 3) {
1240                 print_disk_util(&du->dus, &du->agg, 1, &client->buf);
1241                 __log_buf(&client->buf, "\n");
1242         }
1243 }
1244
1245 static void convert_jobs_eta(struct jobs_eta *je)
1246 {
1247         int i;
1248
1249         je->nr_running          = le32_to_cpu(je->nr_running);
1250         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
1251         je->nr_pending          = le32_to_cpu(je->nr_pending);
1252         je->nr_setting_up       = le32_to_cpu(je->nr_setting_up);
1253         je->files_open          = le32_to_cpu(je->files_open);
1254
1255         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1256                 je->m_rate[i]   = le64_to_cpu(je->m_rate[i]);
1257                 je->t_rate[i]   = le64_to_cpu(je->t_rate[i]);
1258                 je->m_iops[i]   = le32_to_cpu(je->m_iops[i]);
1259                 je->t_iops[i]   = le32_to_cpu(je->t_iops[i]);
1260                 je->rate[i]     = le64_to_cpu(je->rate[i]);
1261                 je->iops[i]     = le32_to_cpu(je->iops[i]);
1262         }
1263
1264         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
1265         je->eta_sec             = le64_to_cpu(je->eta_sec);
1266         je->nr_threads          = le32_to_cpu(je->nr_threads);
1267         je->is_pow2             = le32_to_cpu(je->is_pow2);
1268         je->unit_base           = le32_to_cpu(je->unit_base);
1269         je->sig_figs            = le32_to_cpu(je->sig_figs);
1270 }
1271
1272 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
1273 {
1274         int i;
1275
1276         dst->nr_running         += je->nr_running;
1277         dst->nr_ramp            += je->nr_ramp;
1278         dst->nr_pending         += je->nr_pending;
1279         dst->nr_setting_up      += je->nr_setting_up;
1280         dst->files_open         += je->files_open;
1281
1282         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1283                 dst->m_rate[i]  += je->m_rate[i];
1284                 dst->t_rate[i]  += je->t_rate[i];
1285                 dst->m_iops[i]  += je->m_iops[i];
1286                 dst->t_iops[i]  += je->t_iops[i];
1287                 dst->rate[i]    += je->rate[i];
1288                 dst->iops[i]    += je->iops[i];
1289         }
1290
1291         dst->elapsed_sec        += je->elapsed_sec;
1292
1293         if (je->eta_sec > dst->eta_sec)
1294                 dst->eta_sec = je->eta_sec;
1295
1296         dst->nr_threads         += je->nr_threads;
1297
1298         /*
1299          * This wont be correct for multiple strings, but at least it
1300          * works for the basic cases.
1301          */
1302         strcpy((char *) dst->run_str, (char *) je->run_str);
1303 }
1304
1305 static bool remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
1306 {
1307         struct fio_net_cmd_reply *reply = NULL;
1308         struct flist_head *entry;
1309
1310         flist_for_each(entry, &client->cmd_list) {
1311                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1312
1313                 if (cmd->tag == (uintptr_t) reply)
1314                         break;
1315
1316                 reply = NULL;
1317         }
1318
1319         if (!reply) {
1320                 log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag);
1321                 return false;
1322         }
1323
1324         flist_del(&reply->list);
1325         cmd->tag = reply->saved_tag;
1326         free(reply);
1327         return true;
1328 }
1329
1330 int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
1331 {
1332         do {
1333                 struct fio_net_cmd_reply *reply = NULL;
1334                 struct flist_head *entry;
1335
1336                 flist_for_each(entry, &client->cmd_list) {
1337                         reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1338
1339                         if (tag == (uintptr_t) reply)
1340                                 break;
1341
1342                         reply = NULL;
1343                 }
1344
1345                 if (!reply)
1346                         break;
1347
1348                 usleep(1000);
1349         } while (1);
1350
1351         return 0;
1352 }
1353
1354 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
1355 {
1356         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1357         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
1358
1359         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
1360
1361         assert(client->eta_in_flight == eta);
1362
1363         client->eta_in_flight = NULL;
1364         flist_del_init(&client->eta_list);
1365         client->eta_timeouts = 0;
1366
1367         if (client->ops->jobs_eta)
1368                 client->ops->jobs_eta(client, je);
1369
1370         fio_client_sum_jobs_eta(&eta->eta, je);
1371         fio_client_dec_jobs_eta(eta, client->ops->eta);
1372 }
1373
1374 static void client_flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
1375                                       uint64_t sample_size)
1376 {
1377         struct io_sample *s;
1378         int log_offset;
1379         uint64_t i, j, nr_samples;
1380         struct io_u_plat_entry *entry;
1381         uint64_t *io_u_plat;
1382
1383         int stride = 1 << hist_coarseness;
1384
1385         if (!sample_size)
1386                 return;
1387
1388         s = __get_sample(samples, 0, 0);
1389         log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
1390
1391         nr_samples = sample_size / __log_entry_sz(log_offset);
1392
1393         for (i = 0; i < nr_samples; i++) {
1394
1395                 s = (struct io_sample *)((char *)__get_sample(samples, log_offset, i) +
1396                         i * sizeof(struct io_u_plat_entry));
1397
1398                 entry = s->data.plat_entry;
1399                 io_u_plat = entry->io_u_plat;
1400
1401                 fprintf(f, "%lu, %u, %llu, ", (unsigned long) s->time,
1402                                                 io_sample_ddir(s), (unsigned long long) s->bs);
1403                 for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) {
1404                         fprintf(f, "%llu, ", (unsigned long long)hist_sum(j, stride, io_u_plat, NULL));
1405                 }
1406                 fprintf(f, "%llu\n", (unsigned long long)
1407                         hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat, NULL));
1408
1409         }
1410 }
1411
1412 static int fio_client_handle_iolog(struct fio_client *client,
1413                                    struct fio_net_cmd *cmd)
1414 {
1415         struct cmd_iolog_pdu *pdu = NULL;
1416         bool store_direct;
1417         char *log_pathname = NULL;
1418         int ret = 0;
1419
1420         pdu = convert_iolog(cmd, &store_direct);
1421         if (!pdu) {
1422                 log_err("fio: failed converting IO log\n");
1423                 ret = 1;
1424                 goto out;
1425         }
1426
1427         /* allocate buffer big enough for next sprintf() call */
1428         log_pathname = malloc(10 + strlen((char *)pdu->name) +
1429                         strlen(client->hostname));
1430         if (!log_pathname) {
1431                 log_err("fio: memory allocation of unique pathname failed\n");
1432                 ret = -1;
1433                 goto out;
1434         }
1435         /* generate a unique pathname for the log file using hostname */
1436         sprintf(log_pathname, "%s.%s", pdu->name, client->hostname);
1437
1438         if (store_direct) {
1439                 ssize_t wrote;
1440                 size_t sz;
1441                 int fd;
1442
1443                 fd = open((const char *) log_pathname,
1444                                 O_WRONLY | O_CREAT | O_TRUNC, 0644);
1445                 if (fd < 0) {
1446                         log_err("fio: open log %s: %s\n",
1447                                 log_pathname, strerror(errno));
1448                         ret = 1;
1449                         goto out;
1450                 }
1451
1452                 sz = cmd->pdu_len - sizeof(*pdu);
1453                 wrote = write(fd, pdu->samples, sz);
1454                 close(fd);
1455
1456                 if (wrote != sz) {
1457                         log_err("fio: short write on compressed log\n");
1458                         ret = 1;
1459                         goto out;
1460                 }
1461
1462                 ret = 0;
1463         } else {
1464                 FILE *f;
1465                 f = fopen((const char *) log_pathname, "w");
1466                 if (!f) {
1467                         log_err("fio: fopen log %s : %s\n",
1468                                 log_pathname, strerror(errno));
1469                         ret = 1;
1470                         goto out;
1471                 }
1472
1473                 if (pdu->log_type == IO_LOG_TYPE_HIST) {
1474                         client_flush_hist_samples(f, pdu->log_hist_coarseness, pdu->samples,
1475                                            pdu->nr_samples * sizeof(struct io_sample));
1476                 } else {
1477                         flush_samples(f, pdu->samples,
1478                                         pdu->nr_samples * sizeof(struct io_sample));
1479                 }
1480                 fclose(f);
1481                 ret = 0;
1482         }
1483
1484 out:
1485         if (pdu && pdu != (void *) cmd->payload)
1486                 free(pdu);
1487
1488         if (log_pathname)
1489                 free(log_pathname);
1490
1491         return ret;
1492 }
1493
1494 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
1495 {
1496         struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
1497         const char *os, *arch;
1498         char bit[16];
1499
1500         os = fio_get_os_string(probe->os);
1501         if (!os)
1502                 os = "unknown";
1503
1504         arch = fio_get_arch_string(probe->arch);
1505         if (!arch)
1506                 os = "unknown";
1507
1508         sprintf(bit, "%d-bit", probe->bpp * 8);
1509         probe->flags = le64_to_cpu(probe->flags);
1510
1511         if (output_format & FIO_OUTPUT_NORMAL) {
1512                 log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n",
1513                         probe->hostname, probe->bigendian, bit, os, arch,
1514                         probe->fio_version, (unsigned long) probe->flags);
1515         }
1516
1517         if (!client->name)
1518                 client->name = strdup((char *) probe->hostname);
1519 }
1520
1521 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
1522 {
1523         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1524
1525         client->state = Client_started;
1526         client->jobs = le32_to_cpu(pdu->jobs);
1527         client->nr_stat = le32_to_cpu(pdu->stat_outputs);
1528
1529         if (client->jobs) {
1530                 int i;
1531
1532                 if (client->opt_lists)
1533                         free(client->opt_lists);
1534
1535                 client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
1536                 for (i = 0; i < client->jobs; i++)
1537                         INIT_FLIST_HEAD(&client->opt_lists[i]);
1538         }
1539
1540         sum_stat_clients += client->nr_stat;
1541 }
1542
1543 static void handle_stop(struct fio_client *client)
1544 {
1545         if (client->error)
1546                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
1547 }
1548
1549 static void convert_stop(struct fio_net_cmd *cmd)
1550 {
1551         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1552
1553         pdu->error = le32_to_cpu(pdu->error);
1554 }
1555
1556 static void convert_text(struct fio_net_cmd *cmd)
1557 {
1558         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1559
1560         pdu->level      = le32_to_cpu(pdu->level);
1561         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
1562         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
1563         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
1564 }
1565
1566 static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
1567                                               struct cmd_iolog_pdu *pdu)
1568 {
1569 #ifdef CONFIG_ZLIB
1570         struct cmd_iolog_pdu *ret;
1571         z_stream stream;
1572         uint64_t nr_samples;
1573         size_t total;
1574         char *p;
1575
1576         stream.zalloc = Z_NULL;
1577         stream.zfree = Z_NULL;
1578         stream.opaque = Z_NULL;
1579         stream.avail_in = 0;
1580         stream.next_in = Z_NULL;
1581
1582         if (inflateInit(&stream) != Z_OK)
1583                 return NULL;
1584
1585         /*
1586          * Get header first, it's not compressed
1587          */
1588         nr_samples = le64_to_cpu(pdu->nr_samples);
1589
1590         if (pdu->log_type == IO_LOG_TYPE_HIST)
1591                 total = nr_samples * (__log_entry_sz(le32_to_cpu(pdu->log_offset)) +
1592                                         sizeof(struct io_u_plat_entry));
1593         else
1594                 total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
1595         ret = malloc(total + sizeof(*pdu));
1596         ret->nr_samples = nr_samples;
1597
1598         memcpy(ret, pdu, sizeof(*pdu));
1599
1600         p = (char *) ret + sizeof(*pdu);
1601
1602         stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1603         stream.next_in = (void *)((char *) pdu + sizeof(*pdu));
1604         while (stream.avail_in) {
1605                 unsigned int this_chunk = 65536;
1606                 unsigned int this_len;
1607                 int err;
1608
1609                 if (this_chunk > total)
1610                         this_chunk = total;
1611
1612                 stream.avail_out = this_chunk;
1613                 stream.next_out = (void *)p;
1614                 err = inflate(&stream, Z_NO_FLUSH);
1615                 /* may be Z_OK, or Z_STREAM_END */
1616                 if (err < 0) {
1617                         /*
1618                          * Z_STREAM_ERROR and Z_BUF_ERROR can safely be
1619                          * ignored */
1620                         if (err == Z_STREAM_ERROR || err == Z_BUF_ERROR)
1621                                 break;
1622                         log_err("fio: inflate error %d\n", err);
1623                         free(ret);
1624                         ret = NULL;
1625                         goto err;
1626                 }
1627
1628                 this_len = this_chunk - stream.avail_out;
1629                 p += this_len;
1630                 total -= this_len;
1631         }
1632
1633 err:
1634         inflateEnd(&stream);
1635         return ret;
1636 #else
1637         return NULL;
1638 #endif
1639 }
1640
1641 /*
1642  * This has been compressed on the server side, since it can be big.
1643  * Uncompress here.
1644  */
1645 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd,
1646                                            bool *store_direct)
1647 {
1648         struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1649         struct cmd_iolog_pdu *ret;
1650         uint64_t i;
1651         int compressed;
1652         void *samples;
1653
1654         *store_direct = false;
1655
1656         /*
1657          * Convert if compressed and we support it. If it's not
1658          * compressed, we need not do anything.
1659          */
1660         compressed = le32_to_cpu(pdu->compressed);
1661         if (compressed == XMIT_COMPRESSED) {
1662 #ifndef CONFIG_ZLIB
1663                 log_err("fio: server sent compressed data by mistake\n");
1664                 return NULL;
1665 #endif
1666                 ret = convert_iolog_gz(cmd, pdu);
1667                 if (!ret) {
1668                         log_err("fio: failed decompressing log\n");
1669                         return NULL;
1670                 }
1671         } else if (compressed == STORE_COMPRESSED) {
1672                 *store_direct = true;
1673                 ret = pdu;
1674         } else
1675                 ret = pdu;
1676
1677         ret->nr_samples         = le64_to_cpu(ret->nr_samples);
1678         ret->thread_number      = le32_to_cpu(ret->thread_number);
1679         ret->log_type           = le32_to_cpu(ret->log_type);
1680         ret->compressed         = le32_to_cpu(ret->compressed);
1681         ret->log_offset         = le32_to_cpu(ret->log_offset);
1682         ret->log_hist_coarseness = le32_to_cpu(ret->log_hist_coarseness);
1683
1684         if (*store_direct)
1685                 return ret;
1686
1687         samples = &ret->samples[0];
1688         for (i = 0; i < ret->nr_samples; i++) {
1689                 struct io_sample *s;
1690
1691                 s = __get_sample(samples, ret->log_offset, i);
1692                 if (ret->log_type == IO_LOG_TYPE_HIST)
1693                         s = (struct io_sample *)((char *)s + sizeof(struct io_u_plat_entry) * i);
1694
1695                 s->time         = le64_to_cpu(s->time);
1696                 s->data.val     = le64_to_cpu(s->data.val);
1697                 s->__ddir       = __le32_to_cpu(s->__ddir);
1698                 s->bs           = le64_to_cpu(s->bs);
1699
1700                 if (ret->log_offset) {
1701                         struct io_sample_offset *so = (void *) s;
1702
1703                         so->offset = le64_to_cpu(so->offset);
1704                 }
1705
1706                 if (ret->log_type == IO_LOG_TYPE_HIST) {
1707                         s->data.plat_entry = (struct io_u_plat_entry *)(((char *)s) + sizeof(*s));
1708                         s->data.plat_entry->list.next = NULL;
1709                         s->data.plat_entry->list.prev = NULL;
1710                 }
1711         }
1712
1713         return ret;
1714 }
1715
1716 static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep,
1717                            size_t size, uint64_t tag)
1718 {
1719         rep->error = cpu_to_le32(rep->error);
1720         fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL);
1721 }
1722
1723 static int fio_send_file(struct fio_client *client, struct cmd_sendfile *pdu,
1724                          uint64_t tag)
1725 {
1726         struct cmd_sendfile_reply *rep;
1727         struct stat sb;
1728         size_t size;
1729         int fd;
1730
1731         size = sizeof(*rep);
1732         rep = malloc(size);
1733
1734         if (stat((char *)pdu->path, &sb) < 0) {
1735 fail:
1736                 rep->error = errno;
1737                 sendfile_reply(client->fd, rep, size, tag);
1738                 free(rep);
1739                 return 1;
1740         }
1741
1742         size += sb.st_size;
1743         rep = realloc(rep, size);
1744         rep->size = cpu_to_le32((uint32_t) sb.st_size);
1745
1746         fd = open((char *)pdu->path, O_RDONLY);
1747         if (fd == -1 )
1748                 goto fail;
1749
1750         rep->error = read_data(fd, &rep->data, sb.st_size);
1751         sendfile_reply(client->fd, rep, size, tag);
1752         free(rep);
1753         close(fd);
1754         return 0;
1755 }
1756
1757 int fio_handle_client(struct fio_client *client)
1758 {
1759         struct client_ops *ops = client->ops;
1760         struct fio_net_cmd *cmd;
1761         int size;
1762
1763         dprint(FD_NET, "client: handle %s\n", client->hostname);
1764
1765         cmd = fio_net_recv_cmd(client->fd, true);
1766         if (!cmd)
1767                 return 0;
1768
1769         dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1770                 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1771
1772         client->last_cmd = cmd->opcode;
1773
1774         switch (cmd->opcode) {
1775         case FIO_NET_CMD_QUIT:
1776                 if (ops->quit)
1777                         ops->quit(client, cmd);
1778                 remove_client(client);
1779                 break;
1780         case FIO_NET_CMD_TEXT:
1781                 convert_text(cmd);
1782                 ops->text(client, cmd);
1783                 break;
1784         case FIO_NET_CMD_DU: {
1785                 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1786
1787                 convert_dus(&du->dus);
1788                 convert_agg(&du->agg);
1789
1790                 ops->disk_util(client, cmd);
1791                 break;
1792                 }
1793         case FIO_NET_CMD_TS: {
1794                 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1795
1796                 dprint(FD_NET, "client: ts->ss_state = %u\n", (unsigned int) le32_to_cpu(p->ts.ss_state));
1797                 if (le32_to_cpu(p->ts.ss_state) & FIO_SS_DATA) {
1798                         dprint(FD_NET, "client: received steadystate ring buffers\n");
1799
1800                         size = le64_to_cpu(p->ts.ss_dur);
1801                         p->ts.ss_iops_data = (uint64_t *) ((struct cmd_ts_pdu *)cmd->payload + 1);
1802                         p->ts.ss_bw_data = p->ts.ss_iops_data + size;
1803                 }
1804
1805                 convert_ts(&p->ts, &p->ts);
1806                 convert_gs(&p->rs, &p->rs);
1807
1808                 ops->thread_status(client, cmd);
1809                 break;
1810                 }
1811         case FIO_NET_CMD_GS: {
1812                 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1813
1814                 convert_gs(gs, gs);
1815
1816                 ops->group_stats(client, cmd);
1817                 break;
1818                 }
1819         case FIO_NET_CMD_ETA: {
1820                 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1821
1822                 if (!remove_reply_cmd(client, cmd))
1823                         break;
1824                 convert_jobs_eta(je);
1825                 handle_eta(client, cmd);
1826                 break;
1827                 }
1828         case FIO_NET_CMD_PROBE:
1829                 remove_reply_cmd(client, cmd);
1830                 ops->probe(client, cmd);
1831                 break;
1832         case FIO_NET_CMD_SERVER_START:
1833                 client->state = Client_running;
1834                 if (ops->job_start)
1835                         ops->job_start(client, cmd);
1836                 break;
1837         case FIO_NET_CMD_START: {
1838                 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1839
1840                 pdu->jobs = le32_to_cpu(pdu->jobs);
1841                 ops->start(client, cmd);
1842                 break;
1843                 }
1844         case FIO_NET_CMD_STOP: {
1845                 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1846
1847                 convert_stop(cmd);
1848                 client->state = Client_stopped;
1849                 client->error = le32_to_cpu(pdu->error);
1850                 client->signal = le32_to_cpu(pdu->signal);
1851                 ops->stop(client);
1852                 break;
1853                 }
1854         case FIO_NET_CMD_ADD_JOB: {
1855                 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1856
1857                 client->thread_number = le32_to_cpu(pdu->thread_number);
1858                 client->groupid = le32_to_cpu(pdu->groupid);
1859
1860                 if (ops->add_job)
1861                         ops->add_job(client, cmd);
1862                 break;
1863                 }
1864         case FIO_NET_CMD_IOLOG:
1865                 fio_client_handle_iolog(client, cmd);
1866                 break;
1867         case FIO_NET_CMD_UPDATE_JOB:
1868                 ops->update_job(client, cmd);
1869                 remove_reply_cmd(client, cmd);
1870                 break;
1871         case FIO_NET_CMD_VTRIGGER: {
1872                 struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
1873                 char buf[128];
1874                 int off = 0;
1875
1876                 if (aux_path) {
1877                         strcpy(buf, aux_path);
1878                         off = strlen(buf);
1879                 }
1880
1881                 __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
1882                 exec_trigger(trigger_cmd);
1883                 break;
1884                 }
1885         case FIO_NET_CMD_SENDFILE: {
1886                 struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload;
1887                 fio_send_file(client, pdu, cmd->tag);
1888                 break;
1889                 }
1890         case FIO_NET_CMD_JOB_OPT: {
1891                 handle_job_opt(client, cmd);
1892                 break;
1893         }
1894         default:
1895                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1896                 break;
1897         }
1898
1899         free(cmd);
1900         return 1;
1901 }
1902
1903 int fio_clients_send_trigger(const char *cmd)
1904 {
1905         struct flist_head *entry;
1906         struct fio_client *client;
1907         size_t slen;
1908
1909         dprint(FD_NET, "client: send vtrigger: %s\n", cmd);
1910
1911         if (!cmd)
1912                 slen = 0;
1913         else
1914                 slen = strlen(cmd);
1915
1916         flist_for_each(entry, &client_list) {
1917                 struct cmd_vtrigger_pdu *pdu;
1918
1919                 client = flist_entry(entry, struct fio_client, list);
1920
1921                 pdu = malloc(sizeof(*pdu) + slen);
1922                 pdu->len = cpu_to_le16((uint16_t) slen);
1923                 if (slen)
1924                         memcpy(pdu->cmd, cmd, slen);
1925                 fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu,
1926                                         sizeof(*pdu) + slen, NULL, NULL);
1927                 free(pdu);
1928         }
1929
1930         return 0;
1931 }
1932
1933 static void request_client_etas(struct client_ops *ops)
1934 {
1935         struct fio_client *client;
1936         struct flist_head *entry;
1937         struct client_eta *eta;
1938         int skipped = 0;
1939
1940         if (eta_print == FIO_ETA_NEVER)
1941                 return;
1942
1943         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1944
1945         eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
1946         eta->pending = nr_clients;
1947
1948         flist_for_each(entry, &client_list) {
1949                 client = flist_entry(entry, struct fio_client, list);
1950
1951                 if (!flist_empty(&client->eta_list)) {
1952                         skipped++;
1953                         continue;
1954                 }
1955                 if (client->state != Client_running)
1956                         continue;
1957
1958                 assert(!client->eta_in_flight);
1959                 flist_add_tail(&client->eta_list, &eta_list);
1960                 client->eta_in_flight = eta;
1961                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1962                                         (uintptr_t) eta, &client->cmd_list);
1963         }
1964
1965         while (skipped--) {
1966                 if (!fio_client_dec_jobs_eta(eta, ops->eta))
1967                         break;
1968         }
1969
1970         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1971 }
1972
1973 /*
1974  * A single SEND_ETA timeout isn't fatal. Attempt to recover.
1975  */
1976 static int handle_cmd_timeout(struct fio_client *client,
1977                               struct fio_net_cmd_reply *reply)
1978 {
1979         uint16_t reply_opcode = reply->opcode;
1980
1981         flist_del(&reply->list);
1982         free(reply);
1983
1984         if (reply_opcode != FIO_NET_CMD_SEND_ETA)
1985                 return 1;
1986
1987         log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
1988
1989         flist_del_init(&client->eta_list);
1990         if (client->eta_in_flight) {
1991                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
1992                 client->eta_in_flight = NULL;
1993         }
1994
1995         /*
1996          * If we fail 5 in a row, give up...
1997          */
1998         if (client->eta_timeouts++ > 5)
1999                 return 1;
2000
2001         return 0;
2002 }
2003
2004 static int client_check_cmd_timeout(struct fio_client *client,
2005                                     struct timespec *now)
2006 {
2007         struct fio_net_cmd_reply *reply;
2008         struct flist_head *entry, *tmp;
2009         int ret = 0;
2010
2011         flist_for_each_safe(entry, tmp, &client->cmd_list) {
2012                 unsigned int op;
2013
2014                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
2015
2016                 if (mtime_since(&reply->ts, now) < FIO_NET_CLIENT_TIMEOUT)
2017                         continue;
2018
2019                 op = reply->opcode;
2020                 if (!handle_cmd_timeout(client, reply))
2021                         continue;
2022
2023                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
2024                                                 fio_server_op(op));
2025                 ret = 1;
2026         }
2027
2028         return flist_empty(&client->cmd_list) && ret;
2029 }
2030
2031 static int fio_check_clients_timed_out(void)
2032 {
2033         struct fio_client *client;
2034         struct flist_head *entry, *tmp;
2035         struct timespec ts;
2036         int ret = 0;
2037
2038         fio_gettime(&ts, NULL);
2039
2040         flist_for_each_safe(entry, tmp, &client_list) {
2041                 client = flist_entry(entry, struct fio_client, list);
2042
2043                 if (flist_empty(&client->cmd_list))
2044                         continue;
2045
2046                 if (!client_check_cmd_timeout(client, &ts))
2047                         continue;
2048
2049                 if (client->ops->timed_out)
2050                         client->ops->timed_out(client);
2051                 else
2052                         log_err("fio: client %s timed out\n", client->hostname);
2053
2054                 if (client->last_cmd != FIO_NET_CMD_VTRIGGER)
2055                         client->error = ETIMEDOUT;
2056                 else
2057                         log_info("fio: ignoring timeout due to vtrigger\n");
2058                 remove_client(client);
2059                 ret = 1;
2060         }
2061
2062         return ret;
2063 }
2064
2065 int fio_handle_clients(struct client_ops *ops)
2066 {
2067         struct pollfd *pfds;
2068         int i, ret = 0, retval = 0;
2069
2070         fio_gettime(&eta_ts, NULL);
2071
2072         pfds = malloc(nr_clients * sizeof(struct pollfd));
2073
2074         init_thread_stat(&client_ts);
2075         init_group_run_stat(&client_gs);
2076
2077         while (!exit_backend && nr_clients) {
2078                 struct flist_head *entry, *tmp;
2079                 struct fio_client *client;
2080
2081                 i = 0;
2082                 flist_for_each_safe(entry, tmp, &client_list) {
2083                         client = flist_entry(entry, struct fio_client, list);
2084
2085                         if (!client->sent_job && !client->ops->stay_connected &&
2086                             flist_empty(&client->cmd_list)) {
2087                                 remove_client(client);
2088                                 continue;
2089                         }
2090
2091                         pfds[i].fd = client->fd;
2092                         pfds[i].events = POLLIN;
2093                         i++;
2094                 }
2095
2096                 if (!nr_clients)
2097                         break;
2098
2099                 assert(i == nr_clients);
2100
2101                 do {
2102                         struct timespec ts;
2103                         int timeout;
2104
2105                         fio_gettime(&ts, NULL);
2106                         if (eta_time_within_slack(mtime_since(&eta_ts, &ts))) {
2107                                 request_client_etas(ops);
2108                                 memcpy(&eta_ts, &ts, sizeof(ts));
2109
2110                                 if (fio_check_clients_timed_out())
2111                                         break;
2112                         }
2113
2114                         check_trigger_file();
2115
2116                         timeout = min(100u, ops->eta_msec);
2117
2118                         ret = poll(pfds, nr_clients, timeout);
2119                         if (ret < 0) {
2120                                 if (errno == EINTR)
2121                                         continue;
2122                                 log_err("fio: poll clients: %s\n", strerror(errno));
2123                                 break;
2124                         } else if (!ret)
2125                                 continue;
2126                 } while (ret <= 0);
2127
2128                 for (i = 0; i < nr_clients; i++) {
2129                         if (!(pfds[i].revents & POLLIN))
2130                                 continue;
2131
2132                         client = find_client_by_fd(pfds[i].fd);
2133                         if (!client) {
2134                                 log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd);
2135                                 continue;
2136                         }
2137                         if (!fio_handle_client(client)) {
2138                                 log_info("client: host=%s disconnected\n",
2139                                                 client->hostname);
2140                                 remove_client(client);
2141                                 retval = 1;
2142                         } else if (client->error)
2143                                 retval = 1;
2144                         fio_put_client(client);
2145                 }
2146         }
2147
2148         log_info_buf(allclients.buf, allclients.buflen);
2149         buf_output_free(&allclients);
2150
2151         fio_client_json_fini();
2152
2153         free(pfds);
2154         return retval || error_clients;
2155 }
2156
2157 static void client_display_thread_status(struct jobs_eta *je)
2158 {
2159         if (!(output_format & FIO_OUTPUT_JSON))
2160                 display_thread_status(je);
2161 }