hash: cleanups
[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 = NULL;
288
289                 if (fio_server_poll_fd(client->fd, POLLIN, 0))
290                         cmd = fio_net_recv_cmd(client->fd, false);
291                 if (!cmd)
292                         break;
293
294                 if (cmd->opcode == FIO_NET_CMD_TEXT) {
295                         convert_text(cmd);
296                         client->ops->text(client, cmd);
297                 }
298
299                 free(cmd);
300         } while (1);
301 }
302
303 static void remove_client(struct fio_client *client)
304 {
305         assert(client->refs);
306
307         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
308
309         fio_drain_client_text(client);
310
311         if (!flist_empty(&client->list))
312                 flist_del_init(&client->list);
313
314         fio_client_remove_hash(client);
315
316         if (!flist_empty(&client->eta_list)) {
317                 flist_del_init(&client->eta_list);
318                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
319         }
320
321         close(client->fd);
322         client->fd = -1;
323
324         if (client->ops->removed)
325                 client->ops->removed(client);
326
327         nr_clients--;
328         fio_put_client(client);
329 }
330
331 struct fio_client *fio_get_client(struct fio_client *client)
332 {
333         client->refs++;
334         return client;
335 }
336
337 static void __fio_client_add_cmd_option(struct fio_client *client,
338                                         const char *opt)
339 {
340         int index;
341
342         index = client->argc++;
343         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
344         client->argv[index] = strdup(opt);
345         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
346 }
347
348 void fio_client_add_cmd_option(void *cookie, const char *opt)
349 {
350         struct fio_client *client = cookie;
351         struct flist_head *entry;
352
353         if (!client || !opt)
354                 return;
355
356         __fio_client_add_cmd_option(client, opt);
357
358         /*
359          * Duplicate arguments to shared client group
360          */
361         flist_for_each(entry, &arg_list) {
362                 client = flist_entry(entry, struct fio_client, arg_list);
363
364                 __fio_client_add_cmd_option(client, opt);
365         }
366 }
367
368 static struct fio_client *get_new_client(void)
369 {
370         struct fio_client *client;
371
372         client = malloc(sizeof(*client));
373         memset(client, 0, sizeof(*client));
374
375         INIT_FLIST_HEAD(&client->list);
376         INIT_FLIST_HEAD(&client->hash_list);
377         INIT_FLIST_HEAD(&client->arg_list);
378         INIT_FLIST_HEAD(&client->eta_list);
379         INIT_FLIST_HEAD(&client->cmd_list);
380
381         buf_output_init(&client->buf);
382
383         return client;
384 }
385
386 struct fio_client *fio_client_add_explicit(struct client_ops *ops,
387                                            const char *hostname, int type,
388                                            int port)
389 {
390         struct fio_client *client;
391
392         client = get_new_client();
393
394         if (type == Fio_client_socket)
395                 client->is_sock = true;
396         else {
397                 int ipv6;
398
399                 ipv6 = type == Fio_client_ipv6;
400                 if (fio_server_parse_host(hostname, ipv6,
401                                                 &client->addr.sin_addr,
402                                                 &client->addr6.sin6_addr))
403                         goto err;
404
405                 client->port = port;
406         }
407
408         client->fd = -1;
409         client->ops = ops;
410         client->refs = 1;
411         client->type = ops->client_type;
412         client->hostname = strdup(hostname);
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                 fio_put_client(client);
476                 return -1;
477         }
478
479         client->fd = -1;
480         client->ops = ops;
481         client->refs = 1;
482         client->type = ops->client_type;
483
484         __fio_client_add_cmd_option(client, "fio");
485
486         flist_add(&client->list, &client_list);
487         nr_clients++;
488         dprint(FD_NET, "client: added <%s>\n", client->hostname);
489         *cookie = client;
490         return 0;
491 }
492
493 static const char *server_name(struct fio_client *client, char *buf,
494                                size_t bufsize)
495 {
496         const char *from;
497
498         if (client->ipv6)
499                 from = inet_ntop(AF_INET6, (struct sockaddr *) &client->addr6.sin6_addr, buf, bufsize);
500         else if (client->is_sock)
501                 from = "sock";
502         else
503                 from = inet_ntop(AF_INET, (struct sockaddr *) &client->addr.sin_addr, buf, bufsize);
504
505         return from;
506 }
507
508 static void probe_client(struct fio_client *client)
509 {
510         struct cmd_client_probe_pdu pdu;
511         const char *sname;
512         uint64_t tag;
513         char buf[64];
514
515         dprint(FD_NET, "client: send probe\n");
516
517 #ifdef CONFIG_ZLIB
518         pdu.flags = __le64_to_cpu(FIO_PROBE_FLAG_ZLIB);
519 #else
520         pdu.flags = 0;
521 #endif
522
523         sname = server_name(client, buf, sizeof(buf));
524         memset(pdu.server, 0, sizeof(pdu.server));
525         snprintf((char *) pdu.server, sizeof(pdu.server), "%s", sname);
526
527         fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list);
528 }
529
530 static int fio_client_connect_ip(struct fio_client *client)
531 {
532         struct sockaddr *addr;
533         socklen_t socklen;
534         int fd, domain;
535
536         if (client->ipv6) {
537                 client->addr6.sin6_family = AF_INET6;
538                 client->addr6.sin6_port = htons(client->port);
539                 domain = AF_INET6;
540                 addr = (struct sockaddr *) &client->addr6;
541                 socklen = sizeof(client->addr6);
542         } else {
543                 client->addr.sin_family = AF_INET;
544                 client->addr.sin_port = htons(client->port);
545                 domain = AF_INET;
546                 addr = (struct sockaddr *) &client->addr;
547                 socklen = sizeof(client->addr);
548         }
549
550         fd = socket(domain, SOCK_STREAM, 0);
551         if (fd < 0) {
552                 int ret = -errno;
553
554                 log_err("fio: socket: %s\n", strerror(errno));
555                 return ret;
556         }
557
558         if (connect(fd, addr, socklen) < 0) {
559                 int ret = -errno;
560
561                 log_err("fio: connect: %s\n", strerror(errno));
562                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
563                                                                 client->port);
564                 close(fd);
565                 return ret;
566         }
567
568         return fd;
569 }
570
571 static int fio_client_connect_sock(struct fio_client *client)
572 {
573         struct sockaddr_un *addr = &client->addr_un;
574         socklen_t len;
575         int fd;
576
577         memset(addr, 0, sizeof(*addr));
578         addr->sun_family = AF_UNIX;
579         snprintf(addr->sun_path, sizeof(addr->sun_path), "%s",
580                  client->hostname);
581
582         fd = socket(AF_UNIX, SOCK_STREAM, 0);
583         if (fd < 0) {
584                 int ret = -errno;
585
586                 log_err("fio: socket: %s\n", strerror(errno));
587                 return ret;
588         }
589
590         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
591         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
592                 int ret = -errno;
593
594                 log_err("fio: connect; %s\n", strerror(errno));
595                 close(fd);
596                 return ret;
597         }
598
599         return fd;
600 }
601
602 int fio_client_connect(struct fio_client *client)
603 {
604         int fd;
605
606         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
607
608         if (client->is_sock)
609                 fd = fio_client_connect_sock(client);
610         else
611                 fd = fio_client_connect_ip(client);
612
613         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
614
615         if (fd < 0)
616                 return fd;
617
618         client->fd = fd;
619         fio_client_add_hash(client);
620         client->state = Client_connected;
621
622         probe_client(client);
623         return 0;
624 }
625
626 int fio_client_terminate(struct fio_client *client)
627 {
628         return fio_net_send_quit(client->fd);
629 }
630
631 static void fio_clients_terminate(void)
632 {
633         struct flist_head *entry;
634         struct fio_client *client;
635
636         dprint(FD_NET, "client: terminate clients\n");
637
638         flist_for_each(entry, &client_list) {
639                 client = flist_entry(entry, struct fio_client, list);
640                 fio_client_terminate(client);
641         }
642 }
643
644 static void sig_int(int sig)
645 {
646         dprint(FD_NET, "client: got signal %d\n", sig);
647         fio_clients_terminate();
648 }
649
650 static void client_signal_handler(void)
651 {
652         struct sigaction act;
653
654         memset(&act, 0, sizeof(act));
655         act.sa_handler = sig_int;
656         act.sa_flags = SA_RESTART;
657         sigaction(SIGINT, &act, NULL);
658
659         memset(&act, 0, sizeof(act));
660         act.sa_handler = sig_int;
661         act.sa_flags = SA_RESTART;
662         sigaction(SIGTERM, &act, NULL);
663
664 /* Windows uses SIGBREAK as a quit signal from other applications */
665 #ifdef WIN32
666         memset(&act, 0, sizeof(act));
667         act.sa_handler = sig_int;
668         act.sa_flags = SA_RESTART;
669         sigaction(SIGBREAK, &act, NULL);
670 #endif
671
672         memset(&act, 0, sizeof(act));
673         act.sa_handler = sig_show_status;
674         act.sa_flags = SA_RESTART;
675         sigaction(SIGUSR1, &act, NULL);
676 }
677
678 static int send_client_cmd_line(struct fio_client *client)
679 {
680         struct cmd_single_line_pdu *cslp;
681         struct cmd_line_pdu *clp;
682         unsigned long offset;
683         unsigned int *lens;
684         void *pdu;
685         size_t mem;
686         int i, ret;
687
688         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
689
690         lens = malloc(client->argc * sizeof(unsigned int));
691
692         /*
693          * Find out how much mem we need
694          */
695         for (i = 0, mem = 0; i < client->argc; i++) {
696                 lens[i] = strlen(client->argv[i]) + 1;
697                 mem += lens[i];
698         }
699
700         /*
701          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
702          */
703         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
704
705         pdu = malloc(mem);
706         clp = pdu;
707         offset = sizeof(*clp);
708
709         for (i = 0; i < client->argc; i++) {
710                 uint16_t arg_len = lens[i];
711
712                 cslp = pdu + offset;
713                 strcpy((char *) cslp->text, client->argv[i]);
714                 cslp->len = cpu_to_le16(arg_len);
715                 offset += sizeof(*cslp) + arg_len;
716         }
717
718         free(lens);
719         clp->lines = cpu_to_le16(client->argc);
720         clp->client_type = __cpu_to_le16(client->type);
721         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, NULL, NULL);
722         free(pdu);
723         return ret;
724 }
725
726 int fio_clients_connect(void)
727 {
728         struct fio_client *client;
729         struct flist_head *entry, *tmp;
730         int ret;
731
732 #ifdef WIN32
733         WSADATA wsd;
734         WSAStartup(MAKEWORD(2, 2), &wsd);
735 #endif
736
737         dprint(FD_NET, "client: connect all\n");
738
739         client_signal_handler();
740
741         flist_for_each_safe(entry, tmp, &client_list) {
742                 client = flist_entry(entry, struct fio_client, list);
743
744                 ret = fio_client_connect(client);
745                 if (ret) {
746                         remove_client(client);
747                         continue;
748                 }
749
750                 if (client->argc > 1)
751                         send_client_cmd_line(client);
752         }
753
754         return !nr_clients;
755 }
756
757 int fio_start_client(struct fio_client *client)
758 {
759         dprint(FD_NET, "client: start %s\n", client->hostname);
760         return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
761 }
762
763 int fio_start_all_clients(void)
764 {
765         struct fio_client *client;
766         struct flist_head *entry, *tmp;
767         int ret;
768
769         dprint(FD_NET, "client: start all\n");
770
771         fio_client_json_init();
772
773         flist_for_each_safe(entry, tmp, &client_list) {
774                 client = flist_entry(entry, struct fio_client, list);
775
776                 ret = fio_start_client(client);
777                 if (ret) {
778                         remove_client(client);
779                         continue;
780                 }
781         }
782
783         return flist_empty(&client_list);
784 }
785
786 static int __fio_client_send_remote_ini(struct fio_client *client,
787                                         const char *filename)
788 {
789         struct cmd_load_file_pdu *pdu;
790         size_t p_size;
791         int ret;
792
793         dprint(FD_NET, "send remote ini %s to %s\n", filename, client->hostname);
794
795         p_size = sizeof(*pdu) + strlen(filename) + 1;
796         pdu = malloc(p_size);
797         memset(pdu, 0, p_size);
798         pdu->name_len = strlen(filename);
799         strcpy((char *) pdu->file, filename);
800         pdu->client_type = cpu_to_le16((uint16_t) client->type);
801
802         client->sent_job = true;
803         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_LOAD_FILE, pdu, p_size,NULL, NULL);
804         free(pdu);
805         return ret;
806 }
807
808 /*
809  * Send file contents to server backend. We could use sendfile(), but to remain
810  * more portable lets just read/write the darn thing.
811  */
812 static int __fio_client_send_local_ini(struct fio_client *client,
813                                        const char *filename)
814 {
815         struct cmd_job_pdu *pdu;
816         size_t p_size;
817         struct stat sb;
818         char *p;
819         void *buf;
820         off_t len;
821         int fd, ret;
822
823         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
824
825         fd = open(filename, O_RDONLY);
826         if (fd < 0) {
827                 ret = -errno;
828                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
829                 return ret;
830         }
831
832         if (fstat(fd, &sb) < 0) {
833                 ret = -errno;
834                 log_err("fio: job file stat: %s\n", strerror(errno));
835                 close(fd);
836                 return ret;
837         }
838
839         /*
840          * Add extra space for variable expansion, but doesn't guarantee.
841          */
842         sb.st_size += OPT_LEN_MAX;
843         p_size = sb.st_size + sizeof(*pdu);
844         pdu = malloc(p_size);
845         buf = pdu->buf;
846
847         len = sb.st_size;
848         p = buf;
849         if (read_ini_data(fd, p, len)) {
850                 log_err("fio: failed reading job file %s\n", filename);
851                 close(fd);
852                 free(pdu);
853                 return 1;
854         }
855
856         pdu->buf_len = __cpu_to_le32(sb.st_size);
857         pdu->client_type = cpu_to_le32(client->type);
858
859         client->sent_job = true;
860         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, NULL, NULL);
861         free(pdu);
862         close(fd);
863         return ret;
864 }
865
866 int fio_client_send_ini(struct fio_client *client, const char *filename,
867                         bool remote)
868 {
869         int ret;
870
871         if (!remote)
872                 ret = __fio_client_send_local_ini(client, filename);
873         else
874                 ret = __fio_client_send_remote_ini(client, filename);
875
876         if (!ret)
877                 client->sent_job = true;
878
879         return ret;
880 }
881
882 static int fio_client_send_cf(struct fio_client *client,
883                               struct client_file *cf)
884 {
885         return fio_client_send_ini(client, cf->file, cf->remote);
886 }
887
888 int fio_clients_send_ini(const char *filename)
889 {
890         struct fio_client *client;
891         struct flist_head *entry, *tmp;
892
893         flist_for_each_safe(entry, tmp, &client_list) {
894                 bool failed = false;
895
896                 client = flist_entry(entry, struct fio_client, list);
897
898                 if (client->nr_files) {
899                         int i;
900
901                         for (i = 0; i < client->nr_files; i++) {
902                                 struct client_file *cf;
903
904                                 cf = &client->files[i];
905
906                                 if (fio_client_send_cf(client, cf)) {
907                                         failed = true;
908                                         remove_client(client);
909                                         break;
910                                 }
911                         }
912                 }
913                 if (client->sent_job || failed)
914                         continue;
915                 if (!filename || fio_client_send_ini(client, filename, 0))
916                         remove_client(client);
917         }
918
919         return !nr_clients;
920 }
921
922 int fio_client_update_options(struct fio_client *client,
923                               struct thread_options *o, uint64_t *tag)
924 {
925         struct cmd_add_job_pdu pdu;
926
927         pdu.thread_number = cpu_to_le32(client->thread_number);
928         pdu.groupid = cpu_to_le32(client->groupid);
929         convert_thread_options_to_net(&pdu.top, o);
930
931         return fio_net_send_cmd(client->fd, FIO_NET_CMD_UPDATE_JOB, &pdu, sizeof(pdu), tag, &client->cmd_list);
932 }
933
934 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
935 {
936         dst->max_val    = le64_to_cpu(src->max_val);
937         dst->min_val    = le64_to_cpu(src->min_val);
938         dst->samples    = le64_to_cpu(src->samples);
939
940         /*
941          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
942          */
943         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
944         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
945 }
946
947 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
948 {
949         int i, j, k;
950
951         dst->error              = le32_to_cpu(src->error);
952         dst->thread_number      = le32_to_cpu(src->thread_number);
953         dst->groupid            = le32_to_cpu(src->groupid);
954         dst->pid                = le32_to_cpu(src->pid);
955         dst->members            = le32_to_cpu(src->members);
956         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
957         dst->ioprio             = le32_to_cpu(src->ioprio);
958         dst->disable_prio_stat  = le32_to_cpu(src->disable_prio_stat);
959
960         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
961                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
962                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
963                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
964                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
965                 convert_io_stat(&dst->iops_stat[i], &src->iops_stat[i]);
966         }
967         convert_io_stat(&dst->sync_stat, &src->sync_stat);
968
969         dst->usr_time           = le64_to_cpu(src->usr_time);
970         dst->sys_time           = le64_to_cpu(src->sys_time);
971         dst->ctx                = le64_to_cpu(src->ctx);
972         dst->minf               = le64_to_cpu(src->minf);
973         dst->majf               = le64_to_cpu(src->majf);
974         dst->clat_percentiles   = le32_to_cpu(src->clat_percentiles);
975         dst->lat_percentiles    = le32_to_cpu(src->lat_percentiles);
976         dst->slat_percentiles   = le32_to_cpu(src->slat_percentiles);
977         dst->percentile_precision = le64_to_cpu(src->percentile_precision);
978
979         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
980                 fio_fp64_t *fps = &src->percentile_list[i];
981                 fio_fp64_t *fpd = &dst->percentile_list[i];
982
983                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
984         }
985
986         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
987                 dst->io_u_map[i]        = le64_to_cpu(src->io_u_map[i]);
988                 dst->io_u_submit[i]     = le64_to_cpu(src->io_u_submit[i]);
989                 dst->io_u_complete[i]   = le64_to_cpu(src->io_u_complete[i]);
990         }
991
992         for (i = 0; i < FIO_IO_U_LAT_N_NR; i++)
993                 dst->io_u_lat_n[i]      = le64_to_cpu(src->io_u_lat_n[i]);
994         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
995                 dst->io_u_lat_u[i]      = le64_to_cpu(src->io_u_lat_u[i]);
996         for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
997                 dst->io_u_lat_m[i]      = le64_to_cpu(src->io_u_lat_m[i]);
998
999         for (i = 0; i < FIO_LAT_CNT; i++)
1000                 for (j = 0; j < DDIR_RWDIR_CNT; j++)
1001                         for (k = 0; k < FIO_IO_U_PLAT_NR; k++)
1002                                 dst->io_u_plat[i][j][k] = le64_to_cpu(src->io_u_plat[i][j][k]);
1003
1004         for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
1005                 dst->io_u_sync_plat[j] = le64_to_cpu(src->io_u_sync_plat[j]);
1006
1007         for (i = 0; i < DDIR_RWDIR_SYNC_CNT; i++)
1008                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
1009
1010         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1011                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
1012                 dst->drop_io_u[i]       = le64_to_cpu(src->drop_io_u[i]);
1013         }
1014
1015         dst->total_submit       = le64_to_cpu(src->total_submit);
1016         dst->total_complete     = le64_to_cpu(src->total_complete);
1017         dst->nr_zone_resets     = le64_to_cpu(src->nr_zone_resets);
1018
1019         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1020                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
1021                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
1022         }
1023
1024         dst->total_run_time     = le64_to_cpu(src->total_run_time);
1025         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
1026         dst->total_err_count    = le64_to_cpu(src->total_err_count);
1027         dst->first_error        = le32_to_cpu(src->first_error);
1028         dst->kb_base            = le32_to_cpu(src->kb_base);
1029         dst->unit_base          = le32_to_cpu(src->unit_base);
1030
1031         dst->sig_figs           = le32_to_cpu(src->sig_figs);
1032
1033         dst->latency_depth      = le32_to_cpu(src->latency_depth);
1034         dst->latency_target     = le64_to_cpu(src->latency_target);
1035         dst->latency_window     = le64_to_cpu(src->latency_window);
1036         dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
1037
1038         dst->nr_block_infos     = le64_to_cpu(src->nr_block_infos);
1039         for (i = 0; i < dst->nr_block_infos; i++)
1040                 dst->block_infos[i] = le32_to_cpu(src->block_infos[i]);
1041
1042         dst->ss_dur             = le64_to_cpu(src->ss_dur);
1043         dst->ss_state           = le32_to_cpu(src->ss_state);
1044         dst->ss_head            = le32_to_cpu(src->ss_head);
1045         dst->ss_limit.u.f       = fio_uint64_to_double(le64_to_cpu(src->ss_limit.u.i));
1046         dst->ss_slope.u.f       = fio_uint64_to_double(le64_to_cpu(src->ss_slope.u.i));
1047         dst->ss_deviation.u.f   = fio_uint64_to_double(le64_to_cpu(src->ss_deviation.u.i));
1048         dst->ss_criterion.u.f   = fio_uint64_to_double(le64_to_cpu(src->ss_criterion.u.i));
1049
1050         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1051                 dst->nr_clat_prio[i] = le32_to_cpu(src->nr_clat_prio[i]);
1052                 for (j = 0; j < dst->nr_clat_prio[i]; j++) {
1053                         for (k = 0; k < FIO_IO_U_PLAT_NR; k++)
1054                                 dst->clat_prio[i][j].io_u_plat[k] =
1055                                         le64_to_cpu(src->clat_prio[i][j].io_u_plat[k]);
1056                         convert_io_stat(&dst->clat_prio[i][j].clat_stat,
1057                                         &src->clat_prio[i][j].clat_stat);
1058                         dst->clat_prio[i][j].ioprio =
1059                                 le32_to_cpu(dst->clat_prio[i][j].ioprio);
1060                 }
1061         }
1062
1063         if (dst->ss_state & FIO_SS_DATA) {
1064                 for (i = 0; i < dst->ss_dur; i++ ) {
1065                         dst->ss_iops_data[i] = le64_to_cpu(src->ss_iops_data[i]);
1066                         dst->ss_bw_data[i] = le64_to_cpu(src->ss_bw_data[i]);
1067                 }
1068         }
1069
1070         dst->cachehit           = le64_to_cpu(src->cachehit);
1071         dst->cachemiss          = le64_to_cpu(src->cachemiss);
1072 }
1073
1074 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
1075 {
1076         int i;
1077
1078         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1079                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
1080                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
1081                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
1082                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
1083                 dst->iobytes[i]         = le64_to_cpu(src->iobytes[i]);
1084                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
1085         }
1086
1087         dst->kb_base    = le32_to_cpu(src->kb_base);
1088         dst->unit_base  = le32_to_cpu(src->unit_base);
1089         dst->sig_figs   = le32_to_cpu(src->sig_figs);
1090         dst->groupid    = le32_to_cpu(src->groupid);
1091         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
1092 }
1093
1094 static void json_object_add_client_info(struct json_object *obj,
1095                                         struct fio_client *client)
1096 {
1097         const char *hostname = client->hostname ? client->hostname : "";
1098
1099         json_object_add_value_string(obj, "hostname", hostname);
1100         json_object_add_value_int(obj, "port", client->port);
1101 }
1102
1103 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
1104 {
1105         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1106         struct flist_head *opt_list = NULL;
1107         struct json_object *tsobj;
1108
1109         if (client->opt_lists && p->ts.thread_number <= client->jobs)
1110                 opt_list = &client->opt_lists[p->ts.thread_number - 1];
1111
1112         tsobj = show_thread_status(&p->ts, &p->rs, opt_list, &client->buf);
1113         client->did_stat = true;
1114         if (tsobj) {
1115                 json_object_add_client_info(tsobj, client);
1116                 json_array_add_value_object(clients_array, tsobj);
1117         }
1118
1119         if (sum_stat_clients <= 1)
1120                 return;
1121
1122         sum_thread_stats(&client_ts, &p->ts);
1123         sum_group_stats(&client_gs, &p->rs);
1124
1125         client_ts.members++;
1126         client_ts.thread_number = p->ts.thread_number;
1127         client_ts.groupid = p->ts.groupid;
1128         client_ts.unified_rw_rep = p->ts.unified_rw_rep;
1129         client_ts.sig_figs = p->ts.sig_figs;
1130
1131         if (++sum_stat_nr == sum_stat_clients) {
1132                 strcpy(client_ts.name, "All clients");
1133                 tsobj = show_thread_status(&client_ts, &client_gs, NULL, &allclients);
1134                 if (tsobj) {
1135                         json_object_add_client_info(tsobj, client);
1136                         json_array_add_value_object(clients_array, tsobj);
1137                 }
1138         }
1139 }
1140
1141 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
1142 {
1143         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1144
1145         if (output_format & FIO_OUTPUT_NORMAL)
1146                 show_group_stats(gs, &client->buf);
1147 }
1148
1149 static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
1150 {
1151         struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
1152
1153         pdu->global = le16_to_cpu(pdu->global);
1154         pdu->truncated = le16_to_cpu(pdu->truncated);
1155         pdu->groupid = le32_to_cpu(pdu->groupid);
1156
1157         if (pdu->global) {
1158                 if (!job_opt_object)
1159                         return;
1160
1161                 json_object_add_value_string(job_opt_object,
1162                                              (const char *)pdu->name,
1163                                              (const char *)pdu->value);
1164         } else if (client->opt_lists) {
1165                 struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
1166                 struct print_option *p;
1167
1168                 p = malloc(sizeof(*p));
1169                 p->name = strdup((const char *)pdu->name);
1170                 p->value = pdu->value[0] ? strdup((const char *)pdu->value) :
1171                         NULL;
1172                 flist_add_tail(&p->list, opt_list);
1173         }
1174 }
1175
1176 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
1177 {
1178         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1179         const char *buf = (const char *) pdu->buf;
1180         const char *name;
1181         int fio_unused ret;
1182         struct buf_output out;
1183
1184         buf_output_init(&out);
1185
1186         name = client->name ? client->name : client->hostname;
1187
1188         if (!client->skip_newline && !(output_format & FIO_OUTPUT_TERSE))
1189                 __log_buf(&out, "<%s> ", name);
1190         __log_buf(&out, "%s", buf);
1191         log_info_buf(out.buf, out.buflen);
1192         buf_output_free(&out);
1193         client->skip_newline = strchr(buf, '\n') == NULL;
1194 }
1195
1196 static void convert_agg(struct disk_util_agg *agg)
1197 {
1198         int i;
1199
1200         for (i = 0; i < 2; i++) {
1201                 agg->ios[i]     = le64_to_cpu(agg->ios[i]);
1202                 agg->merges[i]  = le64_to_cpu(agg->merges[i]);
1203                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
1204                 agg->ticks[i]   = le64_to_cpu(agg->ticks[i]);
1205         }
1206
1207         agg->io_ticks           = le64_to_cpu(agg->io_ticks);
1208         agg->time_in_queue      = le64_to_cpu(agg->time_in_queue);
1209         agg->slavecount         = le32_to_cpu(agg->slavecount);
1210         agg->max_util.u.f       = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
1211 }
1212
1213 static void convert_dus(struct disk_util_stat *dus)
1214 {
1215         int i;
1216
1217         for (i = 0; i < 2; i++) {
1218                 dus->s.ios[i]           = le64_to_cpu(dus->s.ios[i]);
1219                 dus->s.merges[i]        = le64_to_cpu(dus->s.merges[i]);
1220                 dus->s.sectors[i]       = le64_to_cpu(dus->s.sectors[i]);
1221                 dus->s.ticks[i]         = le64_to_cpu(dus->s.ticks[i]);
1222         }
1223
1224         dus->s.io_ticks         = le64_to_cpu(dus->s.io_ticks);
1225         dus->s.time_in_queue    = le64_to_cpu(dus->s.time_in_queue);
1226         dus->s.msec             = le64_to_cpu(dus->s.msec);
1227 }
1228
1229 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
1230 {
1231         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1232
1233         if (!client->disk_stats_shown)
1234                 client->disk_stats_shown = true;
1235
1236         if (output_format & FIO_OUTPUT_JSON) {
1237                 struct json_object *duobj;
1238
1239                 json_array_add_disk_util(&du->dus, &du->agg, du_array);
1240                 duobj = json_array_last_value_object(du_array);
1241                 json_object_add_client_info(duobj, client);
1242         }
1243         if (output_format & FIO_OUTPUT_NORMAL) {
1244                 __log_buf(&client->buf, "\nDisk stats (read/write):\n");
1245                 print_disk_util(&du->dus, &du->agg, 0, &client->buf);
1246         }
1247         if (output_format & FIO_OUTPUT_TERSE && terse_version >= 3) {
1248                 print_disk_util(&du->dus, &du->agg, 1, &client->buf);
1249                 __log_buf(&client->buf, "\n");
1250         }
1251 }
1252
1253 static void convert_jobs_eta(struct jobs_eta *je)
1254 {
1255         int i;
1256
1257         je->nr_running          = le32_to_cpu(je->nr_running);
1258         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
1259         je->nr_pending          = le32_to_cpu(je->nr_pending);
1260         je->nr_setting_up       = le32_to_cpu(je->nr_setting_up);
1261         je->files_open          = le32_to_cpu(je->files_open);
1262
1263         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1264                 je->m_rate[i]   = le64_to_cpu(je->m_rate[i]);
1265                 je->t_rate[i]   = le64_to_cpu(je->t_rate[i]);
1266                 je->m_iops[i]   = le32_to_cpu(je->m_iops[i]);
1267                 je->t_iops[i]   = le32_to_cpu(je->t_iops[i]);
1268                 je->rate[i]     = le64_to_cpu(je->rate[i]);
1269                 je->iops[i]     = le32_to_cpu(je->iops[i]);
1270         }
1271
1272         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
1273         je->eta_sec             = le64_to_cpu(je->eta_sec);
1274         je->nr_threads          = le32_to_cpu(je->nr_threads);
1275         je->is_pow2             = le32_to_cpu(je->is_pow2);
1276         je->unit_base           = le32_to_cpu(je->unit_base);
1277         je->sig_figs            = le32_to_cpu(je->sig_figs);
1278 }
1279
1280 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
1281 {
1282         int i;
1283
1284         dst->nr_running         += je->nr_running;
1285         dst->nr_ramp            += je->nr_ramp;
1286         dst->nr_pending         += je->nr_pending;
1287         dst->nr_setting_up      += je->nr_setting_up;
1288         dst->files_open         += je->files_open;
1289
1290         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1291                 dst->m_rate[i]  += je->m_rate[i];
1292                 dst->t_rate[i]  += je->t_rate[i];
1293                 dst->m_iops[i]  += je->m_iops[i];
1294                 dst->t_iops[i]  += je->t_iops[i];
1295                 dst->rate[i]    += je->rate[i];
1296                 dst->iops[i]    += je->iops[i];
1297         }
1298
1299         dst->elapsed_sec        += je->elapsed_sec;
1300
1301         if (je->eta_sec > dst->eta_sec)
1302                 dst->eta_sec = je->eta_sec;
1303
1304         dst->nr_threads         += je->nr_threads;
1305
1306         /*
1307          * This wont be correct for multiple strings, but at least it
1308          * works for the basic cases.
1309          */
1310         strcpy((char *) dst->run_str, (char *) je->run_str);
1311 }
1312
1313 static bool remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
1314 {
1315         struct fio_net_cmd_reply *reply = NULL;
1316         struct flist_head *entry;
1317
1318         flist_for_each(entry, &client->cmd_list) {
1319                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1320
1321                 if (cmd->tag == (uintptr_t) reply)
1322                         break;
1323
1324                 reply = NULL;
1325         }
1326
1327         if (!reply) {
1328                 log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag);
1329                 return false;
1330         }
1331
1332         flist_del(&reply->list);
1333         cmd->tag = reply->saved_tag;
1334         free(reply);
1335         return true;
1336 }
1337
1338 int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
1339 {
1340         do {
1341                 struct fio_net_cmd_reply *reply = NULL;
1342                 struct flist_head *entry;
1343
1344                 flist_for_each(entry, &client->cmd_list) {
1345                         reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1346
1347                         if (tag == (uintptr_t) reply)
1348                                 break;
1349
1350                         reply = NULL;
1351                 }
1352
1353                 if (!reply)
1354                         break;
1355
1356                 usleep(1000);
1357         } while (1);
1358
1359         return 0;
1360 }
1361
1362 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
1363 {
1364         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1365         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
1366
1367         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
1368
1369         assert(client->eta_in_flight == eta);
1370
1371         client->eta_in_flight = NULL;
1372         flist_del_init(&client->eta_list);
1373         client->eta_timeouts = 0;
1374
1375         if (client->ops->jobs_eta)
1376                 client->ops->jobs_eta(client, je);
1377
1378         fio_client_sum_jobs_eta(&eta->eta, je);
1379         fio_client_dec_jobs_eta(eta, client->ops->eta);
1380 }
1381
1382 static void client_flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
1383                                       uint64_t sample_size)
1384 {
1385         struct io_sample *s;
1386         int log_offset;
1387         uint64_t i, j, nr_samples;
1388         struct io_u_plat_entry *entry;
1389         uint64_t *io_u_plat;
1390
1391         int stride = 1 << hist_coarseness;
1392
1393         if (!sample_size)
1394                 return;
1395
1396         s = __get_sample(samples, 0, 0);
1397         log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
1398
1399         nr_samples = sample_size / __log_entry_sz(log_offset);
1400
1401         for (i = 0; i < nr_samples; i++) {
1402
1403                 s = (struct io_sample *)((char *)__get_sample(samples, log_offset, i) +
1404                         i * sizeof(struct io_u_plat_entry));
1405
1406                 entry = s->data.plat_entry;
1407                 io_u_plat = entry->io_u_plat;
1408
1409                 fprintf(f, "%lu, %u, %llu, ", (unsigned long) s->time,
1410                                                 io_sample_ddir(s), (unsigned long long) s->bs);
1411                 for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) {
1412                         fprintf(f, "%llu, ", (unsigned long long)hist_sum(j, stride, io_u_plat, NULL));
1413                 }
1414                 fprintf(f, "%llu\n", (unsigned long long)
1415                         hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat, NULL));
1416
1417         }
1418 }
1419
1420 static int fio_client_handle_iolog(struct fio_client *client,
1421                                    struct fio_net_cmd *cmd)
1422 {
1423         struct cmd_iolog_pdu *pdu = NULL;
1424         bool store_direct;
1425         char *log_pathname = NULL;
1426         int ret = 0;
1427
1428         pdu = convert_iolog(cmd, &store_direct);
1429         if (!pdu) {
1430                 log_err("fio: failed converting IO log\n");
1431                 ret = 1;
1432                 goto out;
1433         }
1434
1435         /* allocate buffer big enough for next sprintf() call */
1436         log_pathname = malloc(10 + strlen((char *)pdu->name) +
1437                         strlen(client->hostname));
1438         if (!log_pathname) {
1439                 log_err("fio: memory allocation of unique pathname failed\n");
1440                 ret = -1;
1441                 goto out;
1442         }
1443         /* generate a unique pathname for the log file using hostname */
1444         sprintf(log_pathname, "%s.%s", pdu->name, client->hostname);
1445
1446         if (store_direct) {
1447                 ssize_t wrote;
1448                 size_t sz;
1449                 int fd;
1450
1451                 fd = open((const char *) log_pathname,
1452                                 O_WRONLY | O_CREAT | O_TRUNC, 0644);
1453                 if (fd < 0) {
1454                         log_err("fio: open log %s: %s\n",
1455                                 log_pathname, strerror(errno));
1456                         ret = 1;
1457                         goto out;
1458                 }
1459
1460                 sz = cmd->pdu_len - sizeof(*pdu);
1461                 wrote = write(fd, pdu->samples, sz);
1462                 close(fd);
1463
1464                 if (wrote != sz) {
1465                         log_err("fio: short write on compressed log\n");
1466                         ret = 1;
1467                         goto out;
1468                 }
1469
1470                 ret = 0;
1471         } else {
1472                 FILE *f;
1473                 f = fopen((const char *) log_pathname, "w");
1474                 if (!f) {
1475                         log_err("fio: fopen log %s : %s\n",
1476                                 log_pathname, strerror(errno));
1477                         ret = 1;
1478                         goto out;
1479                 }
1480
1481                 if (pdu->log_type == IO_LOG_TYPE_HIST) {
1482                         client_flush_hist_samples(f, pdu->log_hist_coarseness, pdu->samples,
1483                                            pdu->nr_samples * sizeof(struct io_sample));
1484                 } else {
1485                         flush_samples(f, pdu->samples,
1486                                         pdu->nr_samples * sizeof(struct io_sample));
1487                 }
1488                 fclose(f);
1489                 ret = 0;
1490         }
1491
1492 out:
1493         if (pdu && pdu != (void *) cmd->payload)
1494                 free(pdu);
1495
1496         if (log_pathname)
1497                 free(log_pathname);
1498
1499         return ret;
1500 }
1501
1502 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
1503 {
1504         struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
1505         const char *os, *arch;
1506         char bit[16];
1507
1508         os = fio_get_os_string(probe->os);
1509         if (!os)
1510                 os = "unknown";
1511
1512         arch = fio_get_arch_string(probe->arch);
1513         if (!arch)
1514                 os = "unknown";
1515
1516         sprintf(bit, "%d-bit", probe->bpp * 8);
1517         probe->flags = le64_to_cpu(probe->flags);
1518
1519         if (output_format & FIO_OUTPUT_NORMAL) {
1520                 log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n",
1521                         probe->hostname, probe->bigendian, bit, os, arch,
1522                         probe->fio_version, (unsigned long) probe->flags);
1523         }
1524
1525         if (!client->name)
1526                 client->name = strdup((char *) probe->hostname);
1527 }
1528
1529 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
1530 {
1531         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1532
1533         client->state = Client_started;
1534         client->jobs = le32_to_cpu(pdu->jobs);
1535         client->nr_stat = le32_to_cpu(pdu->stat_outputs);
1536
1537         if (client->jobs) {
1538                 int i;
1539
1540                 if (client->opt_lists)
1541                         free(client->opt_lists);
1542
1543                 client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
1544                 for (i = 0; i < client->jobs; i++)
1545                         INIT_FLIST_HEAD(&client->opt_lists[i]);
1546         }
1547
1548         sum_stat_clients += client->nr_stat;
1549 }
1550
1551 static void handle_stop(struct fio_client *client)
1552 {
1553         if (client->error)
1554                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
1555 }
1556
1557 static void convert_stop(struct fio_net_cmd *cmd)
1558 {
1559         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1560
1561         pdu->error = le32_to_cpu(pdu->error);
1562 }
1563
1564 static void convert_text(struct fio_net_cmd *cmd)
1565 {
1566         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1567
1568         pdu->level      = le32_to_cpu(pdu->level);
1569         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
1570         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
1571         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
1572 }
1573
1574 static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
1575                                               struct cmd_iolog_pdu *pdu)
1576 {
1577 #ifdef CONFIG_ZLIB
1578         struct cmd_iolog_pdu *ret;
1579         z_stream stream;
1580         uint64_t nr_samples;
1581         size_t total;
1582         char *p;
1583
1584         stream.zalloc = Z_NULL;
1585         stream.zfree = Z_NULL;
1586         stream.opaque = Z_NULL;
1587         stream.avail_in = 0;
1588         stream.next_in = Z_NULL;
1589
1590         if (inflateInit(&stream) != Z_OK)
1591                 return NULL;
1592
1593         /*
1594          * Get header first, it's not compressed
1595          */
1596         nr_samples = le64_to_cpu(pdu->nr_samples);
1597
1598         if (pdu->log_type == IO_LOG_TYPE_HIST)
1599                 total = nr_samples * (__log_entry_sz(le32_to_cpu(pdu->log_offset)) +
1600                                         sizeof(struct io_u_plat_entry));
1601         else
1602                 total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
1603         ret = malloc(total + sizeof(*pdu));
1604         ret->nr_samples = nr_samples;
1605
1606         memcpy(ret, pdu, sizeof(*pdu));
1607
1608         p = (char *) ret + sizeof(*pdu);
1609
1610         stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1611         stream.next_in = (void *)((char *) pdu + sizeof(*pdu));
1612         while (stream.avail_in) {
1613                 unsigned int this_chunk = 65536;
1614                 unsigned int this_len;
1615                 int err;
1616
1617                 if (this_chunk > total)
1618                         this_chunk = total;
1619
1620                 stream.avail_out = this_chunk;
1621                 stream.next_out = (void *)p;
1622                 err = inflate(&stream, Z_NO_FLUSH);
1623                 /* may be Z_OK, or Z_STREAM_END */
1624                 if (err < 0) {
1625                         /*
1626                          * Z_STREAM_ERROR and Z_BUF_ERROR can safely be
1627                          * ignored */
1628                         if (err == Z_STREAM_ERROR || err == Z_BUF_ERROR)
1629                                 break;
1630                         log_err("fio: inflate error %d\n", err);
1631                         free(ret);
1632                         ret = NULL;
1633                         goto err;
1634                 }
1635
1636                 this_len = this_chunk - stream.avail_out;
1637                 p += this_len;
1638                 total -= this_len;
1639         }
1640
1641 err:
1642         inflateEnd(&stream);
1643         return ret;
1644 #else
1645         return NULL;
1646 #endif
1647 }
1648
1649 /*
1650  * This has been compressed on the server side, since it can be big.
1651  * Uncompress here.
1652  */
1653 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd,
1654                                            bool *store_direct)
1655 {
1656         struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1657         struct cmd_iolog_pdu *ret;
1658         uint64_t i;
1659         int compressed;
1660         void *samples;
1661
1662         *store_direct = false;
1663
1664         /*
1665          * Convert if compressed and we support it. If it's not
1666          * compressed, we need not do anything.
1667          */
1668         compressed = le32_to_cpu(pdu->compressed);
1669         if (compressed == XMIT_COMPRESSED) {
1670 #ifndef CONFIG_ZLIB
1671                 log_err("fio: server sent compressed data by mistake\n");
1672                 return NULL;
1673 #endif
1674                 ret = convert_iolog_gz(cmd, pdu);
1675                 if (!ret) {
1676                         log_err("fio: failed decompressing log\n");
1677                         return NULL;
1678                 }
1679         } else if (compressed == STORE_COMPRESSED) {
1680                 *store_direct = true;
1681                 ret = pdu;
1682         } else
1683                 ret = pdu;
1684
1685         ret->nr_samples         = le64_to_cpu(ret->nr_samples);
1686         ret->thread_number      = le32_to_cpu(ret->thread_number);
1687         ret->log_type           = le32_to_cpu(ret->log_type);
1688         ret->compressed         = le32_to_cpu(ret->compressed);
1689         ret->log_offset         = le32_to_cpu(ret->log_offset);
1690         ret->log_prio           = le32_to_cpu(ret->log_prio);
1691         ret->log_hist_coarseness = le32_to_cpu(ret->log_hist_coarseness);
1692
1693         if (*store_direct)
1694                 return ret;
1695
1696         samples = &ret->samples[0];
1697         for (i = 0; i < ret->nr_samples; i++) {
1698                 struct io_sample *s;
1699
1700                 s = __get_sample(samples, ret->log_offset, i);
1701                 if (ret->log_type == IO_LOG_TYPE_HIST)
1702                         s = (struct io_sample *)((char *)s + sizeof(struct io_u_plat_entry) * i);
1703
1704                 s->time         = le64_to_cpu(s->time);
1705                 s->data.val     = le64_to_cpu(s->data.val);
1706                 s->__ddir       = __le32_to_cpu(s->__ddir);
1707                 s->bs           = le64_to_cpu(s->bs);
1708                 s->priority     = le16_to_cpu(s->priority);
1709
1710                 if (ret->log_offset) {
1711                         struct io_sample_offset *so = (void *) s;
1712
1713                         so->offset = le64_to_cpu(so->offset);
1714                 }
1715
1716                 if (ret->log_type == IO_LOG_TYPE_HIST) {
1717                         s->data.plat_entry = (struct io_u_plat_entry *)(((char *)s) + sizeof(*s));
1718                         s->data.plat_entry->list.next = NULL;
1719                         s->data.plat_entry->list.prev = NULL;
1720                 }
1721         }
1722
1723         return ret;
1724 }
1725
1726 static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep,
1727                            size_t size, uint64_t tag)
1728 {
1729         rep->error = cpu_to_le32(rep->error);
1730         fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL);
1731 }
1732
1733 static int fio_send_file(struct fio_client *client, struct cmd_sendfile *pdu,
1734                          uint64_t tag)
1735 {
1736         struct cmd_sendfile_reply *rep;
1737         struct stat sb;
1738         size_t size;
1739         int fd;
1740
1741         size = sizeof(*rep);
1742         rep = malloc(size);
1743
1744         if (stat((char *)pdu->path, &sb) < 0) {
1745 fail:
1746                 rep->error = errno;
1747                 sendfile_reply(client->fd, rep, size, tag);
1748                 free(rep);
1749                 return 1;
1750         }
1751
1752         size += sb.st_size;
1753         rep = realloc(rep, size);
1754         rep->size = cpu_to_le32((uint32_t) sb.st_size);
1755
1756         fd = open((char *)pdu->path, O_RDONLY);
1757         if (fd == -1 )
1758                 goto fail;
1759
1760         rep->error = read_data(fd, &rep->data, sb.st_size);
1761         sendfile_reply(client->fd, rep, size, tag);
1762         free(rep);
1763         close(fd);
1764         return 0;
1765 }
1766
1767 int fio_handle_client(struct fio_client *client)
1768 {
1769         struct client_ops *ops = client->ops;
1770         struct fio_net_cmd *cmd;
1771
1772         dprint(FD_NET, "client: handle %s\n", client->hostname);
1773
1774         cmd = fio_net_recv_cmd(client->fd, true);
1775         if (!cmd)
1776                 return 0;
1777
1778         dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1779                 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1780
1781         client->last_cmd = cmd->opcode;
1782
1783         switch (cmd->opcode) {
1784         case FIO_NET_CMD_QUIT:
1785                 if (ops->quit)
1786                         ops->quit(client, cmd);
1787                 remove_client(client);
1788                 break;
1789         case FIO_NET_CMD_TEXT:
1790                 convert_text(cmd);
1791                 ops->text(client, cmd);
1792                 break;
1793         case FIO_NET_CMD_DU: {
1794                 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1795
1796                 convert_dus(&du->dus);
1797                 convert_agg(&du->agg);
1798
1799                 ops->disk_util(client, cmd);
1800                 break;
1801                 }
1802         case FIO_NET_CMD_TS: {
1803                 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1804                 uint64_t offset;
1805                 int i;
1806
1807                 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1808                         if (le32_to_cpu(p->ts.nr_clat_prio[i])) {
1809                                 offset = le64_to_cpu(p->ts.clat_prio_offset[i]);
1810                                 p->ts.clat_prio[i] =
1811                                         (struct clat_prio_stat *)((char *)p + offset);
1812                         }
1813                 }
1814
1815                 dprint(FD_NET, "client: ts->ss_state = %u\n", (unsigned int) le32_to_cpu(p->ts.ss_state));
1816                 if (le32_to_cpu(p->ts.ss_state) & FIO_SS_DATA) {
1817                         dprint(FD_NET, "client: received steadystate ring buffers\n");
1818
1819                         offset = le64_to_cpu(p->ts.ss_iops_data_offset);
1820                         p->ts.ss_iops_data = (uint64_t *)((char *)p + offset);
1821
1822                         offset = le64_to_cpu(p->ts.ss_bw_data_offset);
1823                         p->ts.ss_bw_data = (uint64_t *)((char *)p + offset);
1824                 }
1825
1826                 convert_ts(&p->ts, &p->ts);
1827                 convert_gs(&p->rs, &p->rs);
1828
1829                 ops->thread_status(client, cmd);
1830                 break;
1831                 }
1832         case FIO_NET_CMD_GS: {
1833                 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1834
1835                 convert_gs(gs, gs);
1836
1837                 ops->group_stats(client, cmd);
1838                 break;
1839                 }
1840         case FIO_NET_CMD_ETA: {
1841                 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1842
1843                 if (!remove_reply_cmd(client, cmd))
1844                         break;
1845                 convert_jobs_eta(je);
1846                 handle_eta(client, cmd);
1847                 break;
1848                 }
1849         case FIO_NET_CMD_PROBE:
1850                 remove_reply_cmd(client, cmd);
1851                 ops->probe(client, cmd);
1852                 break;
1853         case FIO_NET_CMD_SERVER_START:
1854                 client->state = Client_running;
1855                 if (ops->job_start)
1856                         ops->job_start(client, cmd);
1857                 break;
1858         case FIO_NET_CMD_START: {
1859                 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1860
1861                 pdu->jobs = le32_to_cpu(pdu->jobs);
1862                 ops->start(client, cmd);
1863                 break;
1864                 }
1865         case FIO_NET_CMD_STOP: {
1866                 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1867
1868                 convert_stop(cmd);
1869                 client->state = Client_stopped;
1870                 client->error = le32_to_cpu(pdu->error);
1871                 client->signal = le32_to_cpu(pdu->signal);
1872                 ops->stop(client);
1873                 break;
1874                 }
1875         case FIO_NET_CMD_ADD_JOB: {
1876                 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1877
1878                 client->thread_number = le32_to_cpu(pdu->thread_number);
1879                 client->groupid = le32_to_cpu(pdu->groupid);
1880
1881                 if (ops->add_job)
1882                         ops->add_job(client, cmd);
1883                 break;
1884                 }
1885         case FIO_NET_CMD_IOLOG:
1886                 fio_client_handle_iolog(client, cmd);
1887                 break;
1888         case FIO_NET_CMD_UPDATE_JOB:
1889                 ops->update_job(client, cmd);
1890                 remove_reply_cmd(client, cmd);
1891                 break;
1892         case FIO_NET_CMD_VTRIGGER: {
1893                 struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
1894                 char buf[128];
1895                 int off = 0;
1896
1897                 if (aux_path) {
1898                         strcpy(buf, aux_path);
1899                         off = strlen(buf);
1900                 }
1901
1902                 __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
1903                 exec_trigger(trigger_cmd);
1904                 break;
1905                 }
1906         case FIO_NET_CMD_SENDFILE: {
1907                 struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload;
1908                 fio_send_file(client, pdu, cmd->tag);
1909                 break;
1910                 }
1911         case FIO_NET_CMD_JOB_OPT: {
1912                 handle_job_opt(client, cmd);
1913                 break;
1914         }
1915         default:
1916                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1917                 break;
1918         }
1919
1920         free(cmd);
1921         return 1;
1922 }
1923
1924 int fio_clients_send_trigger(const char *cmd)
1925 {
1926         struct flist_head *entry;
1927         struct fio_client *client;
1928         size_t slen;
1929
1930         dprint(FD_NET, "client: send vtrigger: %s\n", cmd);
1931
1932         if (!cmd)
1933                 slen = 0;
1934         else
1935                 slen = strlen(cmd);
1936
1937         flist_for_each(entry, &client_list) {
1938                 struct cmd_vtrigger_pdu *pdu;
1939
1940                 client = flist_entry(entry, struct fio_client, list);
1941
1942                 pdu = malloc(sizeof(*pdu) + slen);
1943                 pdu->len = cpu_to_le16((uint16_t) slen);
1944                 if (slen)
1945                         memcpy(pdu->cmd, cmd, slen);
1946                 fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu,
1947                                         sizeof(*pdu) + slen, NULL, NULL);
1948                 free(pdu);
1949         }
1950
1951         return 0;
1952 }
1953
1954 static void request_client_etas(struct client_ops *ops)
1955 {
1956         struct fio_client *client;
1957         struct flist_head *entry;
1958         struct client_eta *eta;
1959         int skipped = 0;
1960
1961         if (eta_print == FIO_ETA_NEVER)
1962                 return;
1963
1964         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1965
1966         eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
1967         eta->pending = nr_clients;
1968
1969         flist_for_each(entry, &client_list) {
1970                 client = flist_entry(entry, struct fio_client, list);
1971
1972                 if (!flist_empty(&client->eta_list)) {
1973                         skipped++;
1974                         continue;
1975                 }
1976                 if (client->state != Client_running)
1977                         continue;
1978
1979                 assert(!client->eta_in_flight);
1980                 flist_add_tail(&client->eta_list, &eta_list);
1981                 client->eta_in_flight = eta;
1982                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1983                                         (uintptr_t) eta, &client->cmd_list);
1984         }
1985
1986         while (skipped--) {
1987                 if (!fio_client_dec_jobs_eta(eta, ops->eta))
1988                         break;
1989         }
1990
1991         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1992 }
1993
1994 /*
1995  * A single SEND_ETA timeout isn't fatal. Attempt to recover.
1996  */
1997 static int handle_cmd_timeout(struct fio_client *client,
1998                               struct fio_net_cmd_reply *reply)
1999 {
2000         uint16_t reply_opcode = reply->opcode;
2001
2002         flist_del(&reply->list);
2003         free(reply);
2004
2005         if (reply_opcode != FIO_NET_CMD_SEND_ETA)
2006                 return 1;
2007
2008         log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
2009
2010         flist_del_init(&client->eta_list);
2011         if (client->eta_in_flight) {
2012                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
2013                 client->eta_in_flight = NULL;
2014         }
2015
2016         /*
2017          * If we fail 5 in a row, give up...
2018          */
2019         if (client->eta_timeouts++ > 5)
2020                 return 1;
2021
2022         return 0;
2023 }
2024
2025 static int client_check_cmd_timeout(struct fio_client *client,
2026                                     struct timespec *now)
2027 {
2028         struct fio_net_cmd_reply *reply;
2029         struct flist_head *entry, *tmp;
2030         int ret = 0;
2031
2032         flist_for_each_safe(entry, tmp, &client->cmd_list) {
2033                 unsigned int op;
2034
2035                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
2036
2037                 if (mtime_since(&reply->ts, now) < FIO_NET_CLIENT_TIMEOUT)
2038                         continue;
2039
2040                 op = reply->opcode;
2041                 if (!handle_cmd_timeout(client, reply))
2042                         continue;
2043
2044                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
2045                                                 fio_server_op(op));
2046                 ret = 1;
2047         }
2048
2049         return flist_empty(&client->cmd_list) && ret;
2050 }
2051
2052 static int fio_check_clients_timed_out(void)
2053 {
2054         struct fio_client *client;
2055         struct flist_head *entry, *tmp;
2056         struct timespec ts;
2057         int ret = 0;
2058
2059         fio_gettime(&ts, NULL);
2060
2061         flist_for_each_safe(entry, tmp, &client_list) {
2062                 client = flist_entry(entry, struct fio_client, list);
2063
2064                 if (flist_empty(&client->cmd_list))
2065                         continue;
2066
2067                 if (!client_check_cmd_timeout(client, &ts))
2068                         continue;
2069
2070                 if (client->ops->timed_out)
2071                         client->ops->timed_out(client);
2072                 else
2073                         log_err("fio: client %s timed out\n", client->hostname);
2074
2075                 if (client->last_cmd != FIO_NET_CMD_VTRIGGER)
2076                         client->error = ETIMEDOUT;
2077                 else
2078                         log_info("fio: ignoring timeout due to vtrigger\n");
2079                 remove_client(client);
2080                 ret = 1;
2081         }
2082
2083         return ret;
2084 }
2085
2086 int fio_handle_clients(struct client_ops *ops)
2087 {
2088         struct pollfd *pfds;
2089         int i, ret = 0, retval = 0;
2090
2091         fio_gettime(&eta_ts, NULL);
2092
2093         pfds = malloc(nr_clients * sizeof(struct pollfd));
2094
2095         init_thread_stat(&client_ts);
2096         init_group_run_stat(&client_gs);
2097
2098         while (!exit_backend && nr_clients) {
2099                 struct flist_head *entry, *tmp;
2100                 struct fio_client *client;
2101
2102                 i = 0;
2103                 flist_for_each_safe(entry, tmp, &client_list) {
2104                         client = flist_entry(entry, struct fio_client, list);
2105
2106                         if (!client->sent_job && !client->ops->stay_connected &&
2107                             flist_empty(&client->cmd_list)) {
2108                                 remove_client(client);
2109                                 continue;
2110                         }
2111
2112                         pfds[i].fd = client->fd;
2113                         pfds[i].events = POLLIN;
2114                         i++;
2115                 }
2116
2117                 if (!nr_clients)
2118                         break;
2119
2120                 assert(i == nr_clients);
2121
2122                 do {
2123                         struct timespec ts;
2124                         int timeout;
2125
2126                         fio_gettime(&ts, NULL);
2127                         if (eta_time_within_slack(mtime_since(&eta_ts, &ts))) {
2128                                 request_client_etas(ops);
2129                                 memcpy(&eta_ts, &ts, sizeof(ts));
2130
2131                                 if (fio_check_clients_timed_out())
2132                                         break;
2133                         }
2134
2135                         check_trigger_file();
2136
2137                         timeout = min(100u, ops->eta_msec);
2138
2139                         ret = poll(pfds, nr_clients, timeout);
2140                         if (ret < 0) {
2141                                 if (errno == EINTR)
2142                                         continue;
2143                                 log_err("fio: poll clients: %s\n", strerror(errno));
2144                                 break;
2145                         } else if (!ret)
2146                                 continue;
2147                 } while (ret <= 0);
2148
2149                 for (i = 0; i < nr_clients; i++) {
2150                         if (!(pfds[i].revents & POLLIN))
2151                                 continue;
2152
2153                         client = find_client_by_fd(pfds[i].fd);
2154                         if (!client) {
2155                                 log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd);
2156                                 continue;
2157                         }
2158                         if (!fio_handle_client(client)) {
2159                                 log_info("client: host=%s disconnected\n",
2160                                                 client->hostname);
2161                                 remove_client(client);
2162                                 retval = 1;
2163                         } else if (client->error)
2164                                 retval = 1;
2165                         fio_put_client(client);
2166                 }
2167         }
2168
2169         log_info_buf(allclients.buf, allclients.buflen);
2170         buf_output_free(&allclients);
2171
2172         fio_client_json_fini();
2173
2174         free_clat_prio_stats(&client_ts);
2175         free(pfds);
2176         return retval || error_clients;
2177 }
2178
2179 static void client_display_thread_status(struct jobs_eta *je)
2180 {
2181         if (!(output_format & FIO_OUTPUT_JSON))
2182                 display_thread_status(je);
2183 }