Split mutex.c and .h each into three files
[fio.git] / client.c
1 #include <stdio.h>
2 #include <stdlib.h>
3 #include <unistd.h>
4 #include <limits.h>
5 #include <errno.h>
6 #include <fcntl.h>
7 #include <sys/poll.h>
8 #include <sys/types.h>
9 #include <sys/stat.h>
10 #include <sys/wait.h>
11 #include <sys/socket.h>
12 #include <sys/un.h>
13 #include <netinet/in.h>
14 #include <arpa/inet.h>
15 #include <netdb.h>
16 #include <signal.h>
17 #ifdef CONFIG_ZLIB
18 #include <zlib.h>
19 #endif
20
21 #include "fio.h"
22 #include "client.h"
23 #include "server.h"
24 #include "flist.h"
25 #include "hash.h"
26 #include "verify.h"
27
28 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
29 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd);
30 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd);
31 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
32 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd);
33 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd);
34 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd);
35
36 static void convert_text(struct fio_net_cmd *cmd);
37
38 struct client_ops fio_client_ops = {
39         .text           = handle_text,
40         .disk_util      = handle_du,
41         .thread_status  = handle_ts,
42         .group_stats    = handle_gs,
43         .stop           = handle_stop,
44         .start          = handle_start,
45         .eta            = display_thread_status,
46         .probe          = handle_probe,
47         .eta_msec       = FIO_CLIENT_DEF_ETA_MSEC,
48         .client_type    = FIO_CLIENT_TYPE_CLI,
49 };
50
51 static struct timespec eta_ts;
52
53 static FLIST_HEAD(client_list);
54 static FLIST_HEAD(eta_list);
55
56 static FLIST_HEAD(arg_list);
57
58 struct thread_stat client_ts;
59 struct group_run_stats client_gs;
60 int sum_stat_clients;
61
62 static int sum_stat_nr;
63 static struct json_object *root = NULL;
64 static struct json_object *job_opt_object = NULL;
65 static struct json_array *clients_array = NULL;
66 static struct json_array *du_array = NULL;
67
68 static int error_clients;
69
70 #define FIO_CLIENT_HASH_BITS    7
71 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
72 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
73 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
74
75 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *, bool *);
76
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 void fio_client_json_init(void)
124 {
125         char time_buf[32];
126         time_t time_p;
127
128         if (!(output_format & FIO_OUTPUT_JSON))
129                 return;
130
131         time(&time_p);
132         os_ctime_r((const time_t *) &time_p, time_buf, sizeof(time_buf));
133         time_buf[strlen(time_buf) - 1] = '\0';
134
135         root = json_create_object();
136         json_object_add_value_string(root, "fio version", fio_version_string);
137         json_object_add_value_int(root, "timestamp", time_p);
138         json_object_add_value_string(root, "time", time_buf);
139
140         job_opt_object = json_create_object();
141         json_object_add_value_object(root, "global options", job_opt_object);
142         clients_array = json_create_array();
143         json_object_add_value_array(root, "client_stats", clients_array);
144         du_array = json_create_array();
145         json_object_add_value_array(root, "disk_util", du_array);
146 }
147
148 static void fio_client_json_fini(void)
149 {
150         if (!(output_format & FIO_OUTPUT_JSON))
151                 return;
152
153         log_info("\n");
154         json_print_object(root, NULL);
155         log_info("\n");
156         json_free_object(root);
157         root = NULL;
158         clients_array = NULL;
159         du_array = NULL;
160 }
161
162 static struct fio_client *find_client_by_fd(int fd)
163 {
164         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
165         struct fio_client *client;
166         struct flist_head *entry;
167
168         flist_for_each(entry, &client_hash[bucket]) {
169                 client = flist_entry(entry, struct fio_client, hash_list);
170
171                 if (client->fd == fd) {
172                         client->refs++;
173                         return client;
174                 }
175         }
176
177         return NULL;
178 }
179
180 void fio_put_client(struct fio_client *client)
181 {
182         if (--client->refs)
183                 return;
184
185         free(client->hostname);
186         if (client->argv)
187                 free(client->argv);
188         if (client->name)
189                 free(client->name);
190         while (client->nr_files) {
191                 struct client_file *cf = &client->files[--client->nr_files];
192
193                 free(cf->file);
194         }
195         if (client->files)
196                 free(client->files);
197         if (client->opt_lists)
198                 free(client->opt_lists);
199
200         if (!client->did_stat)
201                 sum_stat_clients--;
202
203         if (client->error)
204                 error_clients++;
205
206         free(client);
207 }
208
209 static int fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
210 {
211         if (!--eta->pending) {
212                 eta_fn(&eta->eta);
213                 free(eta);
214                 return 0;
215         }
216
217         return 1;
218 }
219
220 static void fio_drain_client_text(struct fio_client *client)
221 {
222         do {
223                 struct fio_net_cmd *cmd;
224
225                 cmd = fio_net_recv_cmd(client->fd, false);
226                 if (!cmd)
227                         break;
228
229                 if (cmd->opcode == FIO_NET_CMD_TEXT) {
230                         convert_text(cmd);
231                         client->ops->text(client, cmd);
232                 }
233
234                 free(cmd);
235         } while (1);
236 }
237
238 static void remove_client(struct fio_client *client)
239 {
240         assert(client->refs);
241
242         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
243
244         fio_drain_client_text(client);
245
246         if (!flist_empty(&client->list))
247                 flist_del_init(&client->list);
248
249         fio_client_remove_hash(client);
250
251         if (!flist_empty(&client->eta_list)) {
252                 flist_del_init(&client->eta_list);
253                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
254         }
255
256         close(client->fd);
257         client->fd = -1;
258
259         if (client->ops->removed)
260                 client->ops->removed(client);
261
262         nr_clients--;
263         fio_put_client(client);
264 }
265
266 struct fio_client *fio_get_client(struct fio_client *client)
267 {
268         client->refs++;
269         return client;
270 }
271
272 static void __fio_client_add_cmd_option(struct fio_client *client,
273                                         const char *opt)
274 {
275         int index;
276
277         index = client->argc++;
278         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
279         client->argv[index] = strdup(opt);
280         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
281 }
282
283 void fio_client_add_cmd_option(void *cookie, const char *opt)
284 {
285         struct fio_client *client = cookie;
286         struct flist_head *entry;
287
288         if (!client || !opt)
289                 return;
290
291         __fio_client_add_cmd_option(client, opt);
292
293         /*
294          * Duplicate arguments to shared client group
295          */
296         flist_for_each(entry, &arg_list) {
297                 client = flist_entry(entry, struct fio_client, arg_list);
298
299                 __fio_client_add_cmd_option(client, opt);
300         }
301 }
302
303 struct fio_client *fio_client_add_explicit(struct client_ops *ops,
304                                            const char *hostname, int type,
305                                            int port)
306 {
307         struct fio_client *client;
308
309         client = malloc(sizeof(*client));
310         memset(client, 0, sizeof(*client));
311
312         INIT_FLIST_HEAD(&client->list);
313         INIT_FLIST_HEAD(&client->hash_list);
314         INIT_FLIST_HEAD(&client->arg_list);
315         INIT_FLIST_HEAD(&client->eta_list);
316         INIT_FLIST_HEAD(&client->cmd_list);
317
318         client->hostname = strdup(hostname);
319
320         if (type == Fio_client_socket)
321                 client->is_sock = true;
322         else {
323                 int ipv6;
324
325                 ipv6 = type == Fio_client_ipv6;
326                 if (fio_server_parse_host(hostname, ipv6,
327                                                 &client->addr.sin_addr,
328                                                 &client->addr6.sin6_addr))
329                         goto err;
330
331                 client->port = port;
332         }
333
334         client->fd = -1;
335         client->ops = ops;
336         client->refs = 1;
337         client->type = ops->client_type;
338
339         __fio_client_add_cmd_option(client, "fio");
340
341         flist_add(&client->list, &client_list);
342         nr_clients++;
343         dprint(FD_NET, "client: added <%s>\n", client->hostname);
344         return client;
345 err:
346         free(client);
347         return NULL;
348 }
349
350 int fio_client_add_ini_file(void *cookie, const char *ini_file, bool remote)
351 {
352         struct fio_client *client = cookie;
353         struct client_file *cf;
354         size_t new_size;
355         void *new_files;
356
357         if (!client)
358                 return 1;
359
360         dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file);
361
362         new_size = (client->nr_files + 1) * sizeof(struct client_file);
363         new_files = realloc(client->files, new_size);
364         if (!new_files)
365                 return 1;
366
367         client->files = new_files;
368         cf = &client->files[client->nr_files];
369         cf->file = strdup(ini_file);
370         cf->remote = remote;
371         client->nr_files++;
372         return 0;
373 }
374
375 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
376 {
377         struct fio_client *existing = *cookie;
378         struct fio_client *client;
379
380         if (existing) {
381                 /*
382                  * We always add our "exec" name as the option, hence 1
383                  * means empty.
384                  */
385                 if (existing->argc == 1)
386                         flist_add_tail(&existing->arg_list, &arg_list);
387                 else {
388                         while (!flist_empty(&arg_list))
389                                 flist_del_init(arg_list.next);
390                 }
391         }
392
393         client = malloc(sizeof(*client));
394         memset(client, 0, sizeof(*client));
395
396         INIT_FLIST_HEAD(&client->list);
397         INIT_FLIST_HEAD(&client->hash_list);
398         INIT_FLIST_HEAD(&client->arg_list);
399         INIT_FLIST_HEAD(&client->eta_list);
400         INIT_FLIST_HEAD(&client->cmd_list);
401
402         if (fio_server_parse_string(hostname, &client->hostname,
403                                         &client->is_sock, &client->port,
404                                         &client->addr.sin_addr,
405                                         &client->addr6.sin6_addr,
406                                         &client->ipv6))
407                 return -1;
408
409         client->fd = -1;
410         client->ops = ops;
411         client->refs = 1;
412         client->type = ops->client_type;
413
414         __fio_client_add_cmd_option(client, "fio");
415
416         flist_add(&client->list, &client_list);
417         nr_clients++;
418         dprint(FD_NET, "client: added <%s>\n", client->hostname);
419         *cookie = client;
420         return 0;
421 }
422
423 static const char *server_name(struct fio_client *client, char *buf,
424                                size_t bufsize)
425 {
426         const char *from;
427
428         if (client->ipv6)
429                 from = inet_ntop(AF_INET6, (struct sockaddr *) &client->addr6.sin6_addr, buf, bufsize);
430         else if (client->is_sock)
431                 from = "sock";
432         else
433                 from = inet_ntop(AF_INET, (struct sockaddr *) &client->addr.sin_addr, buf, bufsize);
434
435         return from;
436 }
437
438 static void probe_client(struct fio_client *client)
439 {
440         struct cmd_client_probe_pdu pdu;
441         const char *sname;
442         uint64_t tag;
443         char buf[64];
444
445         dprint(FD_NET, "client: send probe\n");
446
447 #ifdef CONFIG_ZLIB
448         pdu.flags = __le64_to_cpu(FIO_PROBE_FLAG_ZLIB);
449 #else
450         pdu.flags = 0;
451 #endif
452
453         sname = server_name(client, buf, sizeof(buf));
454         memset(pdu.server, 0, sizeof(pdu.server));
455         strncpy((char *) pdu.server, sname, sizeof(pdu.server) - 1);
456
457         fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list);
458 }
459
460 static int fio_client_connect_ip(struct fio_client *client)
461 {
462         struct sockaddr *addr;
463         socklen_t socklen;
464         int fd, domain;
465
466         if (client->ipv6) {
467                 client->addr6.sin6_family = AF_INET6;
468                 client->addr6.sin6_port = htons(client->port);
469                 domain = AF_INET6;
470                 addr = (struct sockaddr *) &client->addr6;
471                 socklen = sizeof(client->addr6);
472         } else {
473                 client->addr.sin_family = AF_INET;
474                 client->addr.sin_port = htons(client->port);
475                 domain = AF_INET;
476                 addr = (struct sockaddr *) &client->addr;
477                 socklen = sizeof(client->addr);
478         }
479
480         fd = socket(domain, SOCK_STREAM, 0);
481         if (fd < 0) {
482                 int ret = -errno;
483
484                 log_err("fio: socket: %s\n", strerror(errno));
485                 return ret;
486         }
487
488         if (connect(fd, addr, socklen) < 0) {
489                 int ret = -errno;
490
491                 log_err("fio: connect: %s\n", strerror(errno));
492                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
493                                                                 client->port);
494                 close(fd);
495                 return ret;
496         }
497
498         return fd;
499 }
500
501 static int fio_client_connect_sock(struct fio_client *client)
502 {
503         struct sockaddr_un *addr = &client->addr_un;
504         socklen_t len;
505         int fd;
506
507         memset(addr, 0, sizeof(*addr));
508         addr->sun_family = AF_UNIX;
509         strncpy(addr->sun_path, client->hostname, sizeof(addr->sun_path) - 1);
510
511         fd = socket(AF_UNIX, SOCK_STREAM, 0);
512         if (fd < 0) {
513                 int ret = -errno;
514
515                 log_err("fio: socket: %s\n", strerror(errno));
516                 return ret;
517         }
518
519         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
520         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
521                 int ret = -errno;
522
523                 log_err("fio: connect; %s\n", strerror(errno));
524                 close(fd);
525                 return ret;
526         }
527
528         return fd;
529 }
530
531 int fio_client_connect(struct fio_client *client)
532 {
533         int fd;
534
535         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
536
537         if (client->is_sock)
538                 fd = fio_client_connect_sock(client);
539         else
540                 fd = fio_client_connect_ip(client);
541
542         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
543
544         if (fd < 0)
545                 return fd;
546
547         client->fd = fd;
548         fio_client_add_hash(client);
549         client->state = Client_connected;
550
551         probe_client(client);
552         return 0;
553 }
554
555 int fio_client_terminate(struct fio_client *client)
556 {
557         return fio_net_send_quit(client->fd);
558 }
559
560 static void fio_clients_terminate(void)
561 {
562         struct flist_head *entry;
563         struct fio_client *client;
564
565         dprint(FD_NET, "client: terminate clients\n");
566
567         flist_for_each(entry, &client_list) {
568                 client = flist_entry(entry, struct fio_client, list);
569                 fio_client_terminate(client);
570         }
571 }
572
573 static void sig_int(int sig)
574 {
575         dprint(FD_NET, "client: got signal %d\n", sig);
576         fio_clients_terminate();
577 }
578
579 static void client_signal_handler(void)
580 {
581         struct sigaction act;
582
583         memset(&act, 0, sizeof(act));
584         act.sa_handler = sig_int;
585         act.sa_flags = SA_RESTART;
586         sigaction(SIGINT, &act, NULL);
587
588         memset(&act, 0, sizeof(act));
589         act.sa_handler = sig_int;
590         act.sa_flags = SA_RESTART;
591         sigaction(SIGTERM, &act, NULL);
592
593 /* Windows uses SIGBREAK as a quit signal from other applications */
594 #ifdef WIN32
595         memset(&act, 0, sizeof(act));
596         act.sa_handler = sig_int;
597         act.sa_flags = SA_RESTART;
598         sigaction(SIGBREAK, &act, NULL);
599 #endif
600
601         memset(&act, 0, sizeof(act));
602         act.sa_handler = sig_show_status;
603         act.sa_flags = SA_RESTART;
604         sigaction(SIGUSR1, &act, NULL);
605 }
606
607 static int send_client_cmd_line(struct fio_client *client)
608 {
609         struct cmd_single_line_pdu *cslp;
610         struct cmd_line_pdu *clp;
611         unsigned long offset;
612         unsigned int *lens;
613         void *pdu;
614         size_t mem;
615         int i, ret;
616
617         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
618
619         lens = malloc(client->argc * sizeof(unsigned int));
620
621         /*
622          * Find out how much mem we need
623          */
624         for (i = 0, mem = 0; i < client->argc; i++) {
625                 lens[i] = strlen(client->argv[i]) + 1;
626                 mem += lens[i];
627         }
628
629         /*
630          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
631          */
632         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
633
634         pdu = malloc(mem);
635         clp = pdu;
636         offset = sizeof(*clp);
637
638         for (i = 0; i < client->argc; i++) {
639                 uint16_t arg_len = lens[i];
640
641                 cslp = pdu + offset;
642                 strcpy((char *) cslp->text, client->argv[i]);
643                 cslp->len = cpu_to_le16(arg_len);
644                 offset += sizeof(*cslp) + arg_len;
645         }
646
647         free(lens);
648         clp->lines = cpu_to_le16(client->argc);
649         clp->client_type = __cpu_to_le16(client->type);
650         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, NULL, NULL);
651         free(pdu);
652         return ret;
653 }
654
655 int fio_clients_connect(void)
656 {
657         struct fio_client *client;
658         struct flist_head *entry, *tmp;
659         int ret;
660
661 #ifdef WIN32
662         WSADATA wsd;
663         WSAStartup(MAKEWORD(2, 2), &wsd);
664 #endif
665
666         dprint(FD_NET, "client: connect all\n");
667
668         client_signal_handler();
669
670         flist_for_each_safe(entry, tmp, &client_list) {
671                 client = flist_entry(entry, struct fio_client, list);
672
673                 ret = fio_client_connect(client);
674                 if (ret) {
675                         remove_client(client);
676                         continue;
677                 }
678
679                 if (client->argc > 1)
680                         send_client_cmd_line(client);
681         }
682
683         return !nr_clients;
684 }
685
686 int fio_start_client(struct fio_client *client)
687 {
688         dprint(FD_NET, "client: start %s\n", client->hostname);
689         return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
690 }
691
692 int fio_start_all_clients(void)
693 {
694         struct fio_client *client;
695         struct flist_head *entry, *tmp;
696         int ret;
697
698         dprint(FD_NET, "client: start all\n");
699
700         fio_client_json_init();
701
702         flist_for_each_safe(entry, tmp, &client_list) {
703                 client = flist_entry(entry, struct fio_client, list);
704
705                 ret = fio_start_client(client);
706                 if (ret) {
707                         remove_client(client);
708                         continue;
709                 }
710         }
711
712         return flist_empty(&client_list);
713 }
714
715 static int __fio_client_send_remote_ini(struct fio_client *client,
716                                         const char *filename)
717 {
718         struct cmd_load_file_pdu *pdu;
719         size_t p_size;
720         int ret;
721
722         dprint(FD_NET, "send remote ini %s to %s\n", filename, client->hostname);
723
724         p_size = sizeof(*pdu) + strlen(filename) + 1;
725         pdu = malloc(p_size);
726         memset(pdu, 0, p_size);
727         pdu->name_len = strlen(filename);
728         strcpy((char *) pdu->file, filename);
729         pdu->client_type = cpu_to_le16((uint16_t) client->type);
730
731         client->sent_job = true;
732         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_LOAD_FILE, pdu, p_size,NULL, NULL);
733         free(pdu);
734         return ret;
735 }
736
737 /*
738  * Send file contents to server backend. We could use sendfile(), but to remain
739  * more portable lets just read/write the darn thing.
740  */
741 static int __fio_client_send_local_ini(struct fio_client *client,
742                                        const char *filename)
743 {
744         struct cmd_job_pdu *pdu;
745         size_t p_size;
746         struct stat sb;
747         char *p;
748         void *buf;
749         off_t len;
750         int fd, ret;
751
752         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
753
754         fd = open(filename, O_RDONLY);
755         if (fd < 0) {
756                 ret = -errno;
757                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
758                 return ret;
759         }
760
761         if (fstat(fd, &sb) < 0) {
762                 ret = -errno;
763                 log_err("fio: job file stat: %s\n", strerror(errno));
764                 close(fd);
765                 return ret;
766         }
767
768         p_size = sb.st_size + sizeof(*pdu);
769         pdu = malloc(p_size);
770         buf = pdu->buf;
771
772         len = sb.st_size;
773         p = buf;
774         if (read_data(fd, p, len)) {
775                 log_err("fio: failed reading job file %s\n", filename);
776                 close(fd);
777                 free(pdu);
778                 return 1;
779         }
780
781         pdu->buf_len = __cpu_to_le32(sb.st_size);
782         pdu->client_type = cpu_to_le32(client->type);
783
784         client->sent_job = true;
785         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, NULL, NULL);
786         free(pdu);
787         close(fd);
788         return ret;
789 }
790
791 int fio_client_send_ini(struct fio_client *client, const char *filename,
792                         bool remote)
793 {
794         int ret;
795
796         if (!remote)
797                 ret = __fio_client_send_local_ini(client, filename);
798         else
799                 ret = __fio_client_send_remote_ini(client, filename);
800
801         if (!ret)
802                 client->sent_job = true;
803
804         return ret;
805 }
806
807 static int fio_client_send_cf(struct fio_client *client,
808                               struct client_file *cf)
809 {
810         return fio_client_send_ini(client, cf->file, cf->remote);
811 }
812
813 int fio_clients_send_ini(const char *filename)
814 {
815         struct fio_client *client;
816         struct flist_head *entry, *tmp;
817
818         flist_for_each_safe(entry, tmp, &client_list) {
819                 bool failed = false;
820
821                 client = flist_entry(entry, struct fio_client, list);
822
823                 if (client->nr_files) {
824                         int i;
825
826                         for (i = 0; i < client->nr_files; i++) {
827                                 struct client_file *cf;
828
829                                 cf = &client->files[i];
830
831                                 if (fio_client_send_cf(client, cf)) {
832                                         failed = true;
833                                         remove_client(client);
834                                         break;
835                                 }
836                         }
837                 }
838                 if (client->sent_job || failed)
839                         continue;
840                 if (!filename || fio_client_send_ini(client, filename, 0))
841                         remove_client(client);
842         }
843
844         return !nr_clients;
845 }
846
847 int fio_client_update_options(struct fio_client *client,
848                               struct thread_options *o, uint64_t *tag)
849 {
850         struct cmd_add_job_pdu pdu;
851
852         pdu.thread_number = cpu_to_le32(client->thread_number);
853         pdu.groupid = cpu_to_le32(client->groupid);
854         convert_thread_options_to_net(&pdu.top, o);
855
856         return fio_net_send_cmd(client->fd, FIO_NET_CMD_UPDATE_JOB, &pdu, sizeof(pdu), tag, &client->cmd_list);
857 }
858
859 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
860 {
861         dst->max_val    = le64_to_cpu(src->max_val);
862         dst->min_val    = le64_to_cpu(src->min_val);
863         dst->samples    = le64_to_cpu(src->samples);
864
865         /*
866          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
867          */
868         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
869         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
870 }
871
872 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
873 {
874         int i, j;
875
876         dst->error              = le32_to_cpu(src->error);
877         dst->thread_number      = le32_to_cpu(src->thread_number);
878         dst->groupid            = le32_to_cpu(src->groupid);
879         dst->pid                = le32_to_cpu(src->pid);
880         dst->members            = le32_to_cpu(src->members);
881         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
882
883         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
884                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
885                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
886                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
887                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
888                 convert_io_stat(&dst->iops_stat[i], &src->iops_stat[i]);
889         }
890
891         dst->usr_time           = le64_to_cpu(src->usr_time);
892         dst->sys_time           = le64_to_cpu(src->sys_time);
893         dst->ctx                = le64_to_cpu(src->ctx);
894         dst->minf               = le64_to_cpu(src->minf);
895         dst->majf               = le64_to_cpu(src->majf);
896         dst->clat_percentiles   = le32_to_cpu(src->clat_percentiles);
897         dst->lat_percentiles    = le32_to_cpu(src->lat_percentiles);
898         dst->percentile_precision = le64_to_cpu(src->percentile_precision);
899
900         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
901                 fio_fp64_t *fps = &src->percentile_list[i];
902                 fio_fp64_t *fpd = &dst->percentile_list[i];
903
904                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
905         }
906
907         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
908                 dst->io_u_map[i]        = le64_to_cpu(src->io_u_map[i]);
909                 dst->io_u_submit[i]     = le64_to_cpu(src->io_u_submit[i]);
910                 dst->io_u_complete[i]   = le64_to_cpu(src->io_u_complete[i]);
911         }
912
913         for (i = 0; i < FIO_IO_U_LAT_N_NR; i++)
914                 dst->io_u_lat_n[i]      = le64_to_cpu(src->io_u_lat_n[i]);
915         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
916                 dst->io_u_lat_u[i]      = le64_to_cpu(src->io_u_lat_u[i]);
917         for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
918                 dst->io_u_lat_m[i]      = le64_to_cpu(src->io_u_lat_m[i]);
919
920         for (i = 0; i < DDIR_RWDIR_CNT; i++)
921                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
922                         dst->io_u_plat[i][j] = le64_to_cpu(src->io_u_plat[i][j]);
923
924         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
925                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
926                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
927                 dst->drop_io_u[i]       = le64_to_cpu(src->drop_io_u[i]);
928         }
929
930         dst->total_submit       = le64_to_cpu(src->total_submit);
931         dst->total_complete     = le64_to_cpu(src->total_complete);
932
933         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
934                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
935                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
936         }
937
938         dst->total_run_time     = le64_to_cpu(src->total_run_time);
939         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
940         dst->total_err_count    = le64_to_cpu(src->total_err_count);
941         dst->first_error        = le32_to_cpu(src->first_error);
942         dst->kb_base            = le32_to_cpu(src->kb_base);
943         dst->unit_base          = le32_to_cpu(src->unit_base);
944
945         dst->sig_figs           = le32_to_cpu(src->sig_figs);
946
947         dst->latency_depth      = le32_to_cpu(src->latency_depth);
948         dst->latency_target     = le64_to_cpu(src->latency_target);
949         dst->latency_window     = le64_to_cpu(src->latency_window);
950         dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
951
952         dst->nr_block_infos     = le64_to_cpu(src->nr_block_infos);
953         for (i = 0; i < dst->nr_block_infos; i++)
954                 dst->block_infos[i] = le32_to_cpu(src->block_infos[i]);
955
956         dst->ss_dur             = le64_to_cpu(src->ss_dur);
957         dst->ss_state           = le32_to_cpu(src->ss_state);
958         dst->ss_head            = le32_to_cpu(src->ss_head);
959         dst->ss_limit.u.f       = fio_uint64_to_double(le64_to_cpu(src->ss_limit.u.i));
960         dst->ss_slope.u.f       = fio_uint64_to_double(le64_to_cpu(src->ss_slope.u.i));
961         dst->ss_deviation.u.f   = fio_uint64_to_double(le64_to_cpu(src->ss_deviation.u.i));
962         dst->ss_criterion.u.f   = fio_uint64_to_double(le64_to_cpu(src->ss_criterion.u.i));
963
964         if (dst->ss_state & FIO_SS_DATA) {
965                 for (i = 0; i < dst->ss_dur; i++ ) {
966                         dst->ss_iops_data[i] = le64_to_cpu(src->ss_iops_data[i]);
967                         dst->ss_bw_data[i] = le64_to_cpu(src->ss_bw_data[i]);
968                 }
969         }
970 }
971
972 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
973 {
974         int i;
975
976         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
977                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
978                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
979                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
980                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
981                 dst->iobytes[i]         = le64_to_cpu(src->iobytes[i]);
982                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
983         }
984
985         dst->kb_base    = le32_to_cpu(src->kb_base);
986         dst->unit_base  = le32_to_cpu(src->unit_base);
987         dst->sig_figs   = le32_to_cpu(src->sig_figs);
988         dst->groupid    = le32_to_cpu(src->groupid);
989         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
990 }
991
992 static void json_object_add_client_info(struct json_object *obj,
993                                         struct fio_client *client)
994 {
995         const char *hostname = client->hostname ? client->hostname : "";
996
997         json_object_add_value_string(obj, "hostname", hostname);
998         json_object_add_value_int(obj, "port", client->port);
999 }
1000
1001 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
1002 {
1003         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1004         struct flist_head *opt_list = NULL;
1005         struct json_object *tsobj;
1006
1007         if (client->opt_lists && p->ts.thread_number <= client->jobs)
1008                 opt_list = &client->opt_lists[p->ts.thread_number - 1];
1009
1010         tsobj = show_thread_status(&p->ts, &p->rs, opt_list, NULL);
1011         client->did_stat = true;
1012         if (tsobj) {
1013                 json_object_add_client_info(tsobj, client);
1014                 json_array_add_value_object(clients_array, tsobj);
1015         }
1016
1017         if (sum_stat_clients <= 1)
1018                 return;
1019
1020         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1);
1021         sum_group_stats(&client_gs, &p->rs);
1022
1023         client_ts.members++;
1024         client_ts.thread_number = p->ts.thread_number;
1025         client_ts.groupid = p->ts.groupid;
1026         client_ts.unified_rw_rep = p->ts.unified_rw_rep;
1027         client_ts.sig_figs = p->ts.sig_figs;
1028
1029         if (++sum_stat_nr == sum_stat_clients) {
1030                 strcpy(client_ts.name, "All clients");
1031                 tsobj = show_thread_status(&client_ts, &client_gs, NULL, NULL);
1032                 if (tsobj) {
1033                         json_object_add_client_info(tsobj, client);
1034                         json_array_add_value_object(clients_array, tsobj);
1035                 }
1036         }
1037 }
1038
1039 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
1040 {
1041         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1042
1043         if (output_format & FIO_OUTPUT_NORMAL)
1044                 show_group_stats(gs, NULL);
1045 }
1046
1047 static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
1048 {
1049         struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
1050         struct print_option *p;
1051
1052         if (!job_opt_object)
1053                 return;
1054
1055         pdu->global = le16_to_cpu(pdu->global);
1056         pdu->truncated = le16_to_cpu(pdu->truncated);
1057         pdu->groupid = le32_to_cpu(pdu->groupid);
1058
1059         p = malloc(sizeof(*p));
1060         p->name = strdup((char *) pdu->name);
1061         if (pdu->value[0] != '\0')
1062                 p->value = strdup((char *) pdu->value);
1063         else
1064                 p->value = NULL;
1065
1066         if (pdu->global) {
1067                 const char *pos = "";
1068
1069                 if (p->value)
1070                         pos = p->value;
1071
1072                 json_object_add_value_string(job_opt_object, p->name, pos);
1073         } else if (client->opt_lists) {
1074                 struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
1075
1076                 flist_add_tail(&p->list, opt_list);
1077         }
1078 }
1079
1080 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
1081 {
1082         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1083         const char *buf = (const char *) pdu->buf;
1084         const char *name;
1085         int fio_unused ret;
1086
1087         name = client->name ? client->name : client->hostname;
1088
1089         if (!client->skip_newline)
1090                 fprintf(f_out, "<%s> ", name);
1091         ret = fwrite(buf, pdu->buf_len, 1, f_out);
1092         fflush(f_out);
1093         client->skip_newline = strchr(buf, '\n') == NULL;
1094 }
1095
1096 static void convert_agg(struct disk_util_agg *agg)
1097 {
1098         int i;
1099
1100         for (i = 0; i < 2; i++) {
1101                 agg->ios[i]     = le64_to_cpu(agg->ios[i]);
1102                 agg->merges[i]  = le64_to_cpu(agg->merges[i]);
1103                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
1104                 agg->ticks[i]   = le64_to_cpu(agg->ticks[i]);
1105         }
1106
1107         agg->io_ticks           = le64_to_cpu(agg->io_ticks);
1108         agg->time_in_queue      = le64_to_cpu(agg->time_in_queue);
1109         agg->slavecount         = le32_to_cpu(agg->slavecount);
1110         agg->max_util.u.f       = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
1111 }
1112
1113 static void convert_dus(struct disk_util_stat *dus)
1114 {
1115         int i;
1116
1117         for (i = 0; i < 2; i++) {
1118                 dus->s.ios[i]           = le64_to_cpu(dus->s.ios[i]);
1119                 dus->s.merges[i]        = le64_to_cpu(dus->s.merges[i]);
1120                 dus->s.sectors[i]       = le64_to_cpu(dus->s.sectors[i]);
1121                 dus->s.ticks[i]         = le64_to_cpu(dus->s.ticks[i]);
1122         }
1123
1124         dus->s.io_ticks         = le64_to_cpu(dus->s.io_ticks);
1125         dus->s.time_in_queue    = le64_to_cpu(dus->s.time_in_queue);
1126         dus->s.msec             = le64_to_cpu(dus->s.msec);
1127 }
1128
1129 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
1130 {
1131         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1132
1133         if (!client->disk_stats_shown) {
1134                 client->disk_stats_shown = true;
1135                 log_info("\nDisk stats (read/write):\n");
1136         }
1137
1138         if (output_format & FIO_OUTPUT_JSON) {
1139                 struct json_object *duobj;
1140                 json_array_add_disk_util(&du->dus, &du->agg, du_array);
1141                 duobj = json_array_last_value_object(du_array);
1142                 json_object_add_client_info(duobj, client);
1143         }
1144         if (output_format & FIO_OUTPUT_TERSE)
1145                 print_disk_util(&du->dus, &du->agg, 1, NULL);
1146         if (output_format & FIO_OUTPUT_NORMAL)
1147                 print_disk_util(&du->dus, &du->agg, 0, NULL);
1148 }
1149
1150 static void convert_jobs_eta(struct jobs_eta *je)
1151 {
1152         int i;
1153
1154         je->nr_running          = le32_to_cpu(je->nr_running);
1155         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
1156         je->nr_pending          = le32_to_cpu(je->nr_pending);
1157         je->nr_setting_up       = le32_to_cpu(je->nr_setting_up);
1158         je->files_open          = le32_to_cpu(je->files_open);
1159
1160         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1161                 je->m_rate[i]   = le64_to_cpu(je->m_rate[i]);
1162                 je->t_rate[i]   = le64_to_cpu(je->t_rate[i]);
1163                 je->m_iops[i]   = le32_to_cpu(je->m_iops[i]);
1164                 je->t_iops[i]   = le32_to_cpu(je->t_iops[i]);
1165                 je->rate[i]     = le64_to_cpu(je->rate[i]);
1166                 je->iops[i]     = le32_to_cpu(je->iops[i]);
1167         }
1168
1169         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
1170         je->eta_sec             = le64_to_cpu(je->eta_sec);
1171         je->nr_threads          = le32_to_cpu(je->nr_threads);
1172         je->is_pow2             = le32_to_cpu(je->is_pow2);
1173         je->unit_base           = le32_to_cpu(je->unit_base);
1174         je->sig_figs            = le32_to_cpu(je->sig_figs);
1175 }
1176
1177 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
1178 {
1179         int i;
1180
1181         dst->nr_running         += je->nr_running;
1182         dst->nr_ramp            += je->nr_ramp;
1183         dst->nr_pending         += je->nr_pending;
1184         dst->nr_setting_up      += je->nr_setting_up;
1185         dst->files_open         += je->files_open;
1186
1187         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1188                 dst->m_rate[i]  += je->m_rate[i];
1189                 dst->t_rate[i]  += je->t_rate[i];
1190                 dst->m_iops[i]  += je->m_iops[i];
1191                 dst->t_iops[i]  += je->t_iops[i];
1192                 dst->rate[i]    += je->rate[i];
1193                 dst->iops[i]    += je->iops[i];
1194         }
1195
1196         dst->elapsed_sec        += je->elapsed_sec;
1197
1198         if (je->eta_sec > dst->eta_sec)
1199                 dst->eta_sec = je->eta_sec;
1200
1201         dst->nr_threads         += je->nr_threads;
1202
1203         /*
1204          * This wont be correct for multiple strings, but at least it
1205          * works for the basic cases.
1206          */
1207         strcpy((char *) dst->run_str, (char *) je->run_str);
1208 }
1209
1210 static bool remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
1211 {
1212         struct fio_net_cmd_reply *reply = NULL;
1213         struct flist_head *entry;
1214
1215         flist_for_each(entry, &client->cmd_list) {
1216                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1217
1218                 if (cmd->tag == (uintptr_t) reply)
1219                         break;
1220
1221                 reply = NULL;
1222         }
1223
1224         if (!reply) {
1225                 log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag);
1226                 return false;
1227         }
1228
1229         flist_del(&reply->list);
1230         cmd->tag = reply->saved_tag;
1231         free(reply);
1232         return true;
1233 }
1234
1235 int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
1236 {
1237         do {
1238                 struct fio_net_cmd_reply *reply = NULL;
1239                 struct flist_head *entry;
1240
1241                 flist_for_each(entry, &client->cmd_list) {
1242                         reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1243
1244                         if (tag == (uintptr_t) reply)
1245                                 break;
1246
1247                         reply = NULL;
1248                 }
1249
1250                 if (!reply)
1251                         break;
1252
1253                 usleep(1000);
1254         } while (1);
1255
1256         return 0;
1257 }
1258
1259 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
1260 {
1261         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1262         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
1263
1264         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
1265
1266         assert(client->eta_in_flight == eta);
1267
1268         client->eta_in_flight = NULL;
1269         flist_del_init(&client->eta_list);
1270         client->eta_timeouts = 0;
1271
1272         if (client->ops->jobs_eta)
1273                 client->ops->jobs_eta(client, je);
1274
1275         fio_client_sum_jobs_eta(&eta->eta, je);
1276         fio_client_dec_jobs_eta(eta, client->ops->eta);
1277 }
1278
1279 static void client_flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
1280                                       uint64_t sample_size)
1281 {
1282         struct io_sample *s;
1283         int log_offset;
1284         uint64_t i, j, nr_samples;
1285         struct io_u_plat_entry *entry;
1286         uint64_t *io_u_plat;
1287
1288         int stride = 1 << hist_coarseness;
1289
1290         if (!sample_size)
1291                 return;
1292
1293         s = __get_sample(samples, 0, 0);
1294         log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
1295
1296         nr_samples = sample_size / __log_entry_sz(log_offset);
1297
1298         for (i = 0; i < nr_samples; i++) {
1299
1300                 s = (struct io_sample *)((char *)__get_sample(samples, log_offset, i) +
1301                         i * sizeof(struct io_u_plat_entry));
1302
1303                 entry = s->data.plat_entry;
1304                 io_u_plat = entry->io_u_plat;
1305
1306                 fprintf(f, "%lu, %u, %u, ", (unsigned long) s->time,
1307                                                 io_sample_ddir(s), s->bs);
1308                 for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) {
1309                         fprintf(f, "%llu, ", (unsigned long long)hist_sum(j, stride, io_u_plat, NULL));
1310                 }
1311                 fprintf(f, "%llu\n", (unsigned long long)
1312                         hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat, NULL));
1313
1314         }
1315 }
1316
1317 static int fio_client_handle_iolog(struct fio_client *client,
1318                                    struct fio_net_cmd *cmd)
1319 {
1320         struct cmd_iolog_pdu *pdu = NULL;
1321         bool store_direct;
1322         char *log_pathname = NULL;
1323         int ret = 0;
1324
1325         pdu = convert_iolog(cmd, &store_direct);
1326         if (!pdu) {
1327                 log_err("fio: failed converting IO log\n");
1328                 ret = 1;
1329                 goto out;
1330         }
1331
1332         /* allocate buffer big enough for next sprintf() call */
1333         log_pathname = malloc(10 + strlen((char *)pdu->name) +
1334                         strlen(client->hostname));
1335         if (!log_pathname) {
1336                 log_err("fio: memory allocation of unique pathname failed\n");
1337                 ret = -1;
1338                 goto out;
1339         }
1340         /* generate a unique pathname for the log file using hostname */
1341         sprintf(log_pathname, "%s.%s", pdu->name, client->hostname);
1342
1343         if (store_direct) {
1344                 ssize_t ret;
1345                 size_t sz;
1346                 int fd;
1347
1348                 fd = open((const char *) log_pathname,
1349                                 O_WRONLY | O_CREAT | O_TRUNC, 0644);
1350                 if (fd < 0) {
1351                         log_err("fio: open log %s: %s\n",
1352                                 log_pathname, strerror(errno));
1353                         ret = 1;
1354                         goto out;
1355                 }
1356
1357                 sz = cmd->pdu_len - sizeof(*pdu);
1358                 ret = write(fd, pdu->samples, sz);
1359                 close(fd);
1360
1361                 if (ret != sz) {
1362                         log_err("fio: short write on compressed log\n");
1363                         ret = 1;
1364                         goto out;
1365                 }
1366
1367                 ret = 0;
1368         } else {
1369                 FILE *f;
1370                 f = fopen((const char *) log_pathname, "w");
1371                 if (!f) {
1372                         log_err("fio: fopen log %s : %s\n",
1373                                 log_pathname, strerror(errno));
1374                         ret = 1;
1375                         goto out;
1376                 }
1377
1378                 if (pdu->log_type == IO_LOG_TYPE_HIST) {
1379                         client_flush_hist_samples(f, pdu->log_hist_coarseness, pdu->samples,
1380                                            pdu->nr_samples * sizeof(struct io_sample));
1381                 } else {
1382                         flush_samples(f, pdu->samples,
1383                                         pdu->nr_samples * sizeof(struct io_sample));
1384                 }
1385                 fclose(f);
1386                 ret = 0;
1387         }
1388
1389 out:
1390         if (pdu && pdu != (void *) cmd->payload)
1391                 free(pdu);
1392
1393         if (log_pathname)
1394                 free(log_pathname);
1395
1396         return ret;
1397 }
1398
1399 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
1400 {
1401         struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
1402         const char *os, *arch;
1403         char bit[16];
1404
1405         os = fio_get_os_string(probe->os);
1406         if (!os)
1407                 os = "unknown";
1408
1409         arch = fio_get_arch_string(probe->arch);
1410         if (!arch)
1411                 os = "unknown";
1412
1413         sprintf(bit, "%d-bit", probe->bpp * 8);
1414         probe->flags = le64_to_cpu(probe->flags);
1415
1416         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n",
1417                 probe->hostname, probe->bigendian, bit, os, arch,
1418                 probe->fio_version, (unsigned long) probe->flags);
1419
1420         if (!client->name)
1421                 client->name = strdup((char *) probe->hostname);
1422 }
1423
1424 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
1425 {
1426         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1427
1428         client->state = Client_started;
1429         client->jobs = le32_to_cpu(pdu->jobs);
1430         client->nr_stat = le32_to_cpu(pdu->stat_outputs);
1431
1432         if (client->jobs) {
1433                 int i;
1434
1435                 if (client->opt_lists)
1436                         free(client->opt_lists);
1437
1438                 client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
1439                 for (i = 0; i < client->jobs; i++)
1440                         INIT_FLIST_HEAD(&client->opt_lists[i]);
1441         }
1442
1443         sum_stat_clients += client->nr_stat;
1444 }
1445
1446 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
1447 {
1448         if (client->error)
1449                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
1450 }
1451
1452 static void convert_stop(struct fio_net_cmd *cmd)
1453 {
1454         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1455
1456         pdu->error = le32_to_cpu(pdu->error);
1457 }
1458
1459 static void convert_text(struct fio_net_cmd *cmd)
1460 {
1461         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1462
1463         pdu->level      = le32_to_cpu(pdu->level);
1464         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
1465         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
1466         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
1467 }
1468
1469 static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
1470                                               struct cmd_iolog_pdu *pdu)
1471 {
1472 #ifdef CONFIG_ZLIB
1473         struct cmd_iolog_pdu *ret;
1474         z_stream stream;
1475         uint32_t nr_samples;
1476         size_t total;
1477         char *p;
1478
1479         stream.zalloc = Z_NULL;
1480         stream.zfree = Z_NULL;
1481         stream.opaque = Z_NULL;
1482         stream.avail_in = 0;
1483         stream.next_in = Z_NULL;
1484
1485         if (inflateInit(&stream) != Z_OK)
1486                 return NULL;
1487
1488         /*
1489          * Get header first, it's not compressed
1490          */
1491         nr_samples = le64_to_cpu(pdu->nr_samples);
1492
1493         if (pdu->log_type == IO_LOG_TYPE_HIST)
1494                 total = nr_samples * (__log_entry_sz(le32_to_cpu(pdu->log_offset)) +
1495                                         sizeof(struct io_u_plat_entry));
1496         else
1497                 total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
1498         ret = malloc(total + sizeof(*pdu));
1499         ret->nr_samples = nr_samples;
1500
1501         memcpy(ret, pdu, sizeof(*pdu));
1502
1503         p = (char *) ret + sizeof(*pdu);
1504
1505         stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1506         stream.next_in = (void *)((char *) pdu + sizeof(*pdu));
1507         while (stream.avail_in) {
1508                 unsigned int this_chunk = 65536;
1509                 unsigned int this_len;
1510                 int err;
1511
1512                 if (this_chunk > total)
1513                         this_chunk = total;
1514
1515                 stream.avail_out = this_chunk;
1516                 stream.next_out = (void *)p;
1517                 err = inflate(&stream, Z_NO_FLUSH);
1518                 /* may be Z_OK, or Z_STREAM_END */
1519                 if (err < 0) {
1520                         log_err("fio: inflate error %d\n", err);
1521                         free(ret);
1522                         ret = NULL;
1523                         goto err;
1524                 }
1525
1526                 this_len = this_chunk - stream.avail_out;
1527                 p += this_len;
1528                 total -= this_len;
1529         }
1530
1531 err:
1532         inflateEnd(&stream);
1533         return ret;
1534 #else
1535         return NULL;
1536 #endif
1537 }
1538
1539 /*
1540  * This has been compressed on the server side, since it can be big.
1541  * Uncompress here.
1542  */
1543 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd,
1544                                            bool *store_direct)
1545 {
1546         struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1547         struct cmd_iolog_pdu *ret;
1548         uint64_t i;
1549         int compressed;
1550         void *samples;
1551
1552         *store_direct = false;
1553
1554         /*
1555          * Convert if compressed and we support it. If it's not
1556          * compressed, we need not do anything.
1557          */
1558         compressed = le32_to_cpu(pdu->compressed);
1559         if (compressed == XMIT_COMPRESSED) {
1560 #ifndef CONFIG_ZLIB
1561                 log_err("fio: server sent compressed data by mistake\n");
1562                 return NULL;
1563 #endif
1564                 ret = convert_iolog_gz(cmd, pdu);
1565                 if (!ret) {
1566                         log_err("fio: failed decompressing log\n");
1567                         return NULL;
1568                 }
1569         } else if (compressed == STORE_COMPRESSED) {
1570                 *store_direct = true;
1571                 ret = pdu;
1572         } else
1573                 ret = pdu;
1574
1575         ret->nr_samples         = le64_to_cpu(ret->nr_samples);
1576         ret->thread_number      = le32_to_cpu(ret->thread_number);
1577         ret->log_type           = le32_to_cpu(ret->log_type);
1578         ret->compressed         = le32_to_cpu(ret->compressed);
1579         ret->log_offset         = le32_to_cpu(ret->log_offset);
1580         ret->log_hist_coarseness = le32_to_cpu(ret->log_hist_coarseness);
1581
1582         if (*store_direct)
1583                 return ret;
1584
1585         samples = &ret->samples[0];
1586         for (i = 0; i < ret->nr_samples; i++) {
1587                 struct io_sample *s;
1588
1589                 s = __get_sample(samples, ret->log_offset, i);
1590                 if (ret->log_type == IO_LOG_TYPE_HIST)
1591                         s = (struct io_sample *)((char *)s + sizeof(struct io_u_plat_entry) * i);
1592
1593                 s->time         = le64_to_cpu(s->time);
1594                 s->data.val     = le64_to_cpu(s->data.val);
1595                 s->__ddir       = le32_to_cpu(s->__ddir);
1596                 s->bs           = le32_to_cpu(s->bs);
1597
1598                 if (ret->log_offset) {
1599                         struct io_sample_offset *so = (void *) s;
1600
1601                         so->offset = le64_to_cpu(so->offset);
1602                 }
1603
1604                 if (ret->log_type == IO_LOG_TYPE_HIST) {
1605                         s->data.plat_entry = (struct io_u_plat_entry *)(((char *)s) + sizeof(*s));
1606                         s->data.plat_entry->list.next = NULL;
1607                         s->data.plat_entry->list.prev = NULL;
1608                 }
1609         }
1610
1611         return ret;
1612 }
1613
1614 static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep,
1615                            size_t size, uint64_t tag)
1616 {
1617         rep->error = cpu_to_le32(rep->error);
1618         fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL);
1619 }
1620
1621 static int fio_send_file(struct fio_client *client, struct cmd_sendfile *pdu,
1622                          uint64_t tag)
1623 {
1624         struct cmd_sendfile_reply *rep;
1625         struct stat sb;
1626         size_t size;
1627         int fd;
1628
1629         size = sizeof(*rep);
1630         rep = malloc(size);
1631
1632         if (stat((char *)pdu->path, &sb) < 0) {
1633 fail:
1634                 rep->error = errno;
1635                 sendfile_reply(client->fd, rep, size, tag);
1636                 free(rep);
1637                 return 1;
1638         }
1639
1640         size += sb.st_size;
1641         rep = realloc(rep, size);
1642         rep->size = cpu_to_le32((uint32_t) sb.st_size);
1643
1644         fd = open((char *)pdu->path, O_RDONLY);
1645         if (fd == -1 )
1646                 goto fail;
1647
1648         rep->error = read_data(fd, &rep->data, sb.st_size);
1649         sendfile_reply(client->fd, rep, size, tag);
1650         free(rep);
1651         close(fd);
1652         return 0;
1653 }
1654
1655 int fio_handle_client(struct fio_client *client)
1656 {
1657         struct client_ops *ops = client->ops;
1658         struct fio_net_cmd *cmd;
1659         int size;
1660
1661         dprint(FD_NET, "client: handle %s\n", client->hostname);
1662
1663         cmd = fio_net_recv_cmd(client->fd, true);
1664         if (!cmd)
1665                 return 0;
1666
1667         dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1668                 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1669
1670         client->last_cmd = cmd->opcode;
1671
1672         switch (cmd->opcode) {
1673         case FIO_NET_CMD_QUIT:
1674                 if (ops->quit)
1675                         ops->quit(client, cmd);
1676                 remove_client(client);
1677                 break;
1678         case FIO_NET_CMD_TEXT:
1679                 convert_text(cmd);
1680                 ops->text(client, cmd);
1681                 break;
1682         case FIO_NET_CMD_DU: {
1683                 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1684
1685                 convert_dus(&du->dus);
1686                 convert_agg(&du->agg);
1687
1688                 ops->disk_util(client, cmd);
1689                 break;
1690                 }
1691         case FIO_NET_CMD_TS: {
1692                 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1693
1694                 dprint(FD_NET, "client: ts->ss_state = %u\n", (unsigned int) le32_to_cpu(p->ts.ss_state));
1695                 if (le32_to_cpu(p->ts.ss_state) & FIO_SS_DATA) {
1696                         dprint(FD_NET, "client: received steadystate ring buffers\n");
1697
1698                         size = le64_to_cpu(p->ts.ss_dur);
1699                         p->ts.ss_iops_data = (uint64_t *) ((struct cmd_ts_pdu *)cmd->payload + 1);
1700                         p->ts.ss_bw_data = p->ts.ss_iops_data + size;
1701                 }
1702
1703                 convert_ts(&p->ts, &p->ts);
1704                 convert_gs(&p->rs, &p->rs);
1705
1706                 ops->thread_status(client, cmd);
1707                 break;
1708                 }
1709         case FIO_NET_CMD_GS: {
1710                 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1711
1712                 convert_gs(gs, gs);
1713
1714                 ops->group_stats(client, cmd);
1715                 break;
1716                 }
1717         case FIO_NET_CMD_ETA: {
1718                 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1719
1720                 if (!remove_reply_cmd(client, cmd))
1721                         break;
1722                 convert_jobs_eta(je);
1723                 handle_eta(client, cmd);
1724                 break;
1725                 }
1726         case FIO_NET_CMD_PROBE:
1727                 remove_reply_cmd(client, cmd);
1728                 ops->probe(client, cmd);
1729                 break;
1730         case FIO_NET_CMD_SERVER_START:
1731                 client->state = Client_running;
1732                 if (ops->job_start)
1733                         ops->job_start(client, cmd);
1734                 break;
1735         case FIO_NET_CMD_START: {
1736                 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1737
1738                 pdu->jobs = le32_to_cpu(pdu->jobs);
1739                 ops->start(client, cmd);
1740                 break;
1741                 }
1742         case FIO_NET_CMD_STOP: {
1743                 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1744
1745                 convert_stop(cmd);
1746                 client->state = Client_stopped;
1747                 client->error = le32_to_cpu(pdu->error);
1748                 client->signal = le32_to_cpu(pdu->signal);
1749                 ops->stop(client, cmd);
1750                 break;
1751                 }
1752         case FIO_NET_CMD_ADD_JOB: {
1753                 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1754
1755                 client->thread_number = le32_to_cpu(pdu->thread_number);
1756                 client->groupid = le32_to_cpu(pdu->groupid);
1757
1758                 if (ops->add_job)
1759                         ops->add_job(client, cmd);
1760                 break;
1761                 }
1762         case FIO_NET_CMD_IOLOG:
1763                 fio_client_handle_iolog(client, cmd);
1764                 break;
1765         case FIO_NET_CMD_UPDATE_JOB:
1766                 ops->update_job(client, cmd);
1767                 remove_reply_cmd(client, cmd);
1768                 break;
1769         case FIO_NET_CMD_VTRIGGER: {
1770                 struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
1771                 char buf[128];
1772                 int off = 0;
1773
1774                 if (aux_path) {
1775                         strcpy(buf, aux_path);
1776                         off = strlen(buf);
1777                 }
1778
1779                 __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
1780                 exec_trigger(trigger_cmd);
1781                 break;
1782                 }
1783         case FIO_NET_CMD_SENDFILE: {
1784                 struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload;
1785                 fio_send_file(client, pdu, cmd->tag);
1786                 break;
1787                 }
1788         case FIO_NET_CMD_JOB_OPT: {
1789                 handle_job_opt(client, cmd);
1790                 break;
1791         }
1792         default:
1793                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1794                 break;
1795         }
1796
1797         free(cmd);
1798         return 1;
1799 }
1800
1801 int fio_clients_send_trigger(const char *cmd)
1802 {
1803         struct flist_head *entry;
1804         struct fio_client *client;
1805         size_t slen;
1806
1807         dprint(FD_NET, "client: send vtrigger: %s\n", cmd);
1808
1809         if (!cmd)
1810                 slen = 0;
1811         else
1812                 slen = strlen(cmd);
1813
1814         flist_for_each(entry, &client_list) {
1815                 struct cmd_vtrigger_pdu *pdu;
1816
1817                 client = flist_entry(entry, struct fio_client, list);
1818
1819                 pdu = malloc(sizeof(*pdu) + slen);
1820                 pdu->len = cpu_to_le16((uint16_t) slen);
1821                 if (slen)
1822                         memcpy(pdu->cmd, cmd, slen);
1823                 fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu,
1824                                         sizeof(*pdu) + slen, NULL, NULL);
1825                 free(pdu);
1826         }
1827
1828         return 0;
1829 }
1830
1831 static void request_client_etas(struct client_ops *ops)
1832 {
1833         struct fio_client *client;
1834         struct flist_head *entry;
1835         struct client_eta *eta;
1836         int skipped = 0;
1837
1838         if (eta_print == FIO_ETA_NEVER)
1839                 return;
1840
1841         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1842
1843         eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
1844         eta->pending = nr_clients;
1845
1846         flist_for_each(entry, &client_list) {
1847                 client = flist_entry(entry, struct fio_client, list);
1848
1849                 if (!flist_empty(&client->eta_list)) {
1850                         skipped++;
1851                         continue;
1852                 }
1853                 if (client->state != Client_running)
1854                         continue;
1855
1856                 assert(!client->eta_in_flight);
1857                 flist_add_tail(&client->eta_list, &eta_list);
1858                 client->eta_in_flight = eta;
1859                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1860                                         (uintptr_t) eta, &client->cmd_list);
1861         }
1862
1863         while (skipped--) {
1864                 if (!fio_client_dec_jobs_eta(eta, ops->eta))
1865                         break;
1866         }
1867
1868         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1869 }
1870
1871 /*
1872  * A single SEND_ETA timeout isn't fatal. Attempt to recover.
1873  */
1874 static int handle_cmd_timeout(struct fio_client *client,
1875                               struct fio_net_cmd_reply *reply)
1876 {
1877         uint16_t reply_opcode = reply->opcode;
1878
1879         flist_del(&reply->list);
1880         free(reply);
1881
1882         if (reply_opcode != FIO_NET_CMD_SEND_ETA)
1883                 return 1;
1884
1885         log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
1886
1887         flist_del_init(&client->eta_list);
1888         if (client->eta_in_flight) {
1889                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
1890                 client->eta_in_flight = NULL;
1891         }
1892
1893         /*
1894          * If we fail 5 in a row, give up...
1895          */
1896         if (client->eta_timeouts++ > 5)
1897                 return 1;
1898
1899         return 0;
1900 }
1901
1902 static int client_check_cmd_timeout(struct fio_client *client,
1903                                     struct timespec *now)
1904 {
1905         struct fio_net_cmd_reply *reply;
1906         struct flist_head *entry, *tmp;
1907         int ret = 0;
1908
1909         flist_for_each_safe(entry, tmp, &client->cmd_list) {
1910                 unsigned int op;
1911
1912                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1913
1914                 if (mtime_since(&reply->ts, now) < FIO_NET_CLIENT_TIMEOUT)
1915                         continue;
1916
1917                 op = reply->opcode;
1918                 if (!handle_cmd_timeout(client, reply))
1919                         continue;
1920
1921                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1922                                                 fio_server_op(op));
1923                 ret = 1;
1924         }
1925
1926         return flist_empty(&client->cmd_list) && ret;
1927 }
1928
1929 static int fio_check_clients_timed_out(void)
1930 {
1931         struct fio_client *client;
1932         struct flist_head *entry, *tmp;
1933         struct timespec ts;
1934         int ret = 0;
1935
1936         fio_gettime(&ts, NULL);
1937
1938         flist_for_each_safe(entry, tmp, &client_list) {
1939                 client = flist_entry(entry, struct fio_client, list);
1940
1941                 if (flist_empty(&client->cmd_list))
1942                         continue;
1943
1944                 if (!client_check_cmd_timeout(client, &ts))
1945                         continue;
1946
1947                 if (client->ops->timed_out)
1948                         client->ops->timed_out(client);
1949                 else
1950                         log_err("fio: client %s timed out\n", client->hostname);
1951
1952                 if (client->last_cmd != FIO_NET_CMD_VTRIGGER)
1953                         client->error = ETIMEDOUT;
1954                 else
1955                         log_info("fio: ignoring timeout due to vtrigger\n");
1956                 remove_client(client);
1957                 ret = 1;
1958         }
1959
1960         return ret;
1961 }
1962
1963 int fio_handle_clients(struct client_ops *ops)
1964 {
1965         struct pollfd *pfds;
1966         int i, ret = 0, retval = 0;
1967
1968         fio_gettime(&eta_ts, NULL);
1969
1970         pfds = malloc(nr_clients * sizeof(struct pollfd));
1971
1972         init_thread_stat(&client_ts);
1973         init_group_run_stat(&client_gs);
1974
1975         while (!exit_backend && nr_clients) {
1976                 struct flist_head *entry, *tmp;
1977                 struct fio_client *client;
1978
1979                 i = 0;
1980                 flist_for_each_safe(entry, tmp, &client_list) {
1981                         client = flist_entry(entry, struct fio_client, list);
1982
1983                         if (!client->sent_job && !client->ops->stay_connected &&
1984                             flist_empty(&client->cmd_list)) {
1985                                 remove_client(client);
1986                                 continue;
1987                         }
1988
1989                         pfds[i].fd = client->fd;
1990                         pfds[i].events = POLLIN;
1991                         i++;
1992                 }
1993
1994                 if (!nr_clients)
1995                         break;
1996
1997                 assert(i == nr_clients);
1998
1999                 do {
2000                         struct timespec ts;
2001                         int timeout;
2002
2003                         fio_gettime(&ts, NULL);
2004                         if (eta_time_within_slack(mtime_since(&eta_ts, &ts))) {
2005                                 request_client_etas(ops);
2006                                 memcpy(&eta_ts, &ts, sizeof(ts));
2007
2008                                 if (fio_check_clients_timed_out())
2009                                         break;
2010                         }
2011
2012                         check_trigger_file();
2013
2014                         timeout = min(100u, ops->eta_msec);
2015
2016                         ret = poll(pfds, nr_clients, timeout);
2017                         if (ret < 0) {
2018                                 if (errno == EINTR)
2019                                         continue;
2020                                 log_err("fio: poll clients: %s\n", strerror(errno));
2021                                 break;
2022                         } else if (!ret)
2023                                 continue;
2024                 } while (ret <= 0);
2025
2026                 for (i = 0; i < nr_clients; i++) {
2027                         if (!(pfds[i].revents & POLLIN))
2028                                 continue;
2029
2030                         client = find_client_by_fd(pfds[i].fd);
2031                         if (!client) {
2032                                 log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd);
2033                                 continue;
2034                         }
2035                         if (!fio_handle_client(client)) {
2036                                 log_info("client: host=%s disconnected\n",
2037                                                 client->hostname);
2038                                 remove_client(client);
2039                                 retval = 1;
2040                         } else if (client->error)
2041                                 retval = 1;
2042                         fio_put_client(client);
2043                 }
2044         }
2045
2046         fio_client_json_fini();
2047
2048         free(pfds);
2049         return retval || error_clients;
2050 }