Fio 2.0.4
[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
18 #include "fio.h"
19 #include "server.h"
20 #include "flist.h"
21 #include "hash.h"
22
23 struct client_eta {
24         struct jobs_eta eta;
25         unsigned int pending;
26 };
27
28 struct fio_client {
29         struct flist_head list;
30         struct flist_head hash_list;
31         struct flist_head arg_list;
32         union {
33                 struct sockaddr_in addr;
34                 struct sockaddr_in6 addr6;
35                 struct sockaddr_un addr_un;
36         };
37         char *hostname;
38         int port;
39         int fd;
40
41         char *name;
42
43         int state;
44
45         int skip_newline;
46         int is_sock;
47         int disk_stats_shown;
48         unsigned int jobs;
49         int error;
50         int ipv6;
51         int sent_job;
52
53         struct flist_head eta_list;
54         struct client_eta *eta_in_flight;
55
56         struct flist_head cmd_list;
57
58         uint16_t argc;
59         char **argv;
60 };
61
62 static struct timeval eta_tv;
63
64 enum {
65         Client_created          = 0,
66         Client_connected        = 1,
67         Client_started          = 2,
68         Client_running          = 3,
69         Client_stopped          = 4,
70         Client_exited           = 5,
71 };
72
73 static FLIST_HEAD(client_list);
74 static FLIST_HEAD(eta_list);
75
76 static FLIST_HEAD(arg_list);
77
78 static struct thread_stat client_ts;
79 static struct group_run_stats client_gs;
80 static int sum_stat_clients;
81 static int sum_stat_nr;
82
83 #define FIO_CLIENT_HASH_BITS    7
84 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
85 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
86 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
87
88 static int handle_client(struct fio_client *client);
89 static void dec_jobs_eta(struct client_eta *eta);
90
91 static void fio_client_add_hash(struct fio_client *client)
92 {
93         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
94
95         bucket &= FIO_CLIENT_HASH_MASK;
96         flist_add(&client->hash_list, &client_hash[bucket]);
97 }
98
99 static void fio_client_remove_hash(struct fio_client *client)
100 {
101         if (!flist_empty(&client->hash_list))
102                 flist_del_init(&client->hash_list);
103 }
104
105 static void fio_init fio_client_hash_init(void)
106 {
107         int i;
108
109         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
110                 INIT_FLIST_HEAD(&client_hash[i]);
111 }
112
113 static struct fio_client *find_client_by_fd(int fd)
114 {
115         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
116         struct fio_client *client;
117         struct flist_head *entry;
118
119         flist_for_each(entry, &client_hash[bucket]) {
120                 client = flist_entry(entry, struct fio_client, hash_list);
121
122                 if (client->fd == fd)
123                         return client;
124         }
125
126         return NULL;
127 }
128
129 static void remove_client(struct fio_client *client)
130 {
131         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
132         flist_del(&client->list);
133
134         fio_client_remove_hash(client);
135
136         if (!flist_empty(&client->eta_list)) {
137                 flist_del_init(&client->eta_list);
138                 dec_jobs_eta(client->eta_in_flight);
139         }
140
141         free(client->hostname);
142         if (client->argv)
143                 free(client->argv);
144         if (client->name)
145                 free(client->name);
146
147         free(client);
148         nr_clients--;
149         sum_stat_clients--;
150 }
151
152 static void __fio_client_add_cmd_option(struct fio_client *client,
153                                         const char *opt)
154 {
155         int index;
156
157         index = client->argc++;
158         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
159         client->argv[index] = strdup(opt);
160         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
161 }
162
163 void fio_client_add_cmd_option(void *cookie, const char *opt)
164 {
165         struct fio_client *client = cookie;
166         struct flist_head *entry;
167
168         if (!client || !opt)
169                 return;
170
171         __fio_client_add_cmd_option(client, opt);
172
173         /*
174          * Duplicate arguments to shared client group
175          */
176         flist_for_each(entry, &arg_list) {
177                 client = flist_entry(entry, struct fio_client, arg_list);
178
179                 __fio_client_add_cmd_option(client, opt);
180         }
181 }
182
183 int fio_client_add(const char *hostname, void **cookie)
184 {
185         struct fio_client *existing = *cookie;
186         struct fio_client *client;
187
188         if (existing) {
189                 /*
190                  * We always add our "exec" name as the option, hence 1
191                  * means empty.
192                  */
193                 if (existing->argc == 1)
194                         flist_add_tail(&existing->arg_list, &arg_list);
195                 else {
196                         while (!flist_empty(&arg_list))
197                                 flist_del_init(arg_list.next);
198                 }
199         }
200
201         client = malloc(sizeof(*client));
202         memset(client, 0, sizeof(*client));
203
204         INIT_FLIST_HEAD(&client->list);
205         INIT_FLIST_HEAD(&client->hash_list);
206         INIT_FLIST_HEAD(&client->arg_list);
207         INIT_FLIST_HEAD(&client->eta_list);
208         INIT_FLIST_HEAD(&client->cmd_list);
209
210         if (fio_server_parse_string(hostname, &client->hostname,
211                                         &client->is_sock, &client->port,
212                                         &client->addr.sin_addr,
213                                         &client->addr6.sin6_addr,
214                                         &client->ipv6))
215                 return -1;
216
217         client->fd = -1;
218
219         __fio_client_add_cmd_option(client, "fio");
220
221         flist_add(&client->list, &client_list);
222         nr_clients++;
223         dprint(FD_NET, "client: added <%s>\n", client->hostname);
224         *cookie = client;
225         return 0;
226 }
227
228 static int fio_client_connect_ip(struct fio_client *client)
229 {
230         struct sockaddr *addr;
231         fio_socklen_t socklen;
232         int fd, domain;
233
234         if (client->ipv6) {
235                 client->addr6.sin6_family = AF_INET6;
236                 client->addr6.sin6_port = htons(client->port);
237                 domain = AF_INET6;
238                 addr = (struct sockaddr *) &client->addr6;
239                 socklen = sizeof(client->addr6);
240         } else {
241                 client->addr.sin_family = AF_INET;
242                 client->addr.sin_port = htons(client->port);
243                 domain = AF_INET;
244                 addr = (struct sockaddr *) &client->addr;
245                 socklen = sizeof(client->addr);
246         }
247
248         fd = socket(domain, SOCK_STREAM, 0);
249         if (fd < 0) {
250                 log_err("fio: socket: %s\n", strerror(errno));
251                 return -1;
252         }
253
254         if (connect(fd, addr, socklen) < 0) {
255                 log_err("fio: connect: %s\n", strerror(errno));
256                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
257                                                                 client->port);
258                 close(fd);
259                 return -1;
260         }
261
262         return fd;
263 }
264
265 static int fio_client_connect_sock(struct fio_client *client)
266 {
267         struct sockaddr_un *addr = &client->addr_un;
268         fio_socklen_t len;
269         int fd;
270
271         memset(addr, 0, sizeof(*addr));
272         addr->sun_family = AF_UNIX;
273         strcpy(addr->sun_path, client->hostname);
274
275         fd = socket(AF_UNIX, SOCK_STREAM, 0);
276         if (fd < 0) {
277                 log_err("fio: socket: %s\n", strerror(errno));
278                 return -1;
279         }
280
281         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
282         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
283                 log_err("fio: connect; %s\n", strerror(errno));
284                 close(fd);
285                 return -1;
286         }
287
288         return fd;
289 }
290
291 static int fio_client_connect(struct fio_client *client)
292 {
293         int fd;
294
295         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
296
297         if (client->is_sock)
298                 fd = fio_client_connect_sock(client);
299         else
300                 fd = fio_client_connect_ip(client);
301
302         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
303
304         if (fd < 0)
305                 return 1;
306
307         client->fd = fd;
308         fio_client_add_hash(client);
309         client->state = Client_connected;
310         return 0;
311 }
312
313 void fio_clients_terminate(void)
314 {
315         struct flist_head *entry;
316         struct fio_client *client;
317
318         dprint(FD_NET, "client: terminate clients\n");
319
320         flist_for_each(entry, &client_list) {
321                 client = flist_entry(entry, struct fio_client, list);
322
323                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0, NULL);
324         }
325 }
326
327 static void sig_int(int sig)
328 {
329         dprint(FD_NET, "client: got signal %d\n", sig);
330         fio_clients_terminate();
331 }
332
333 static void client_signal_handler(void)
334 {
335         struct sigaction act;
336
337         memset(&act, 0, sizeof(act));
338         act.sa_handler = sig_int;
339         act.sa_flags = SA_RESTART;
340         sigaction(SIGINT, &act, NULL);
341
342         memset(&act, 0, sizeof(act));
343         act.sa_handler = sig_int;
344         act.sa_flags = SA_RESTART;
345         sigaction(SIGTERM, &act, NULL);
346 }
347
348 static void probe_client(struct fio_client *client)
349 {
350         dprint(FD_NET, "client: send probe\n");
351
352         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
353 }
354
355 static int send_client_cmd_line(struct fio_client *client)
356 {
357         struct cmd_single_line_pdu *cslp;
358         struct cmd_line_pdu *clp;
359         unsigned long offset;
360         unsigned int *lens;
361         void *pdu;
362         size_t mem;
363         int i, ret;
364
365         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
366
367         lens = malloc(client->argc * sizeof(unsigned int));
368
369         /*
370          * Find out how much mem we need
371          */
372         for (i = 0, mem = 0; i < client->argc; i++) {
373                 lens[i] = strlen(client->argv[i]) + 1;
374                 mem += lens[i];
375         }
376
377         /*
378          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
379          */
380         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
381
382         pdu = malloc(mem);
383         clp = pdu;
384         offset = sizeof(*clp);
385
386         for (i = 0; i < client->argc; i++) {
387                 uint16_t arg_len = lens[i];
388
389                 cslp = pdu + offset;
390                 strcpy((char *) cslp->text, client->argv[i]);
391                 cslp->len = cpu_to_le16(arg_len);
392                 offset += sizeof(*cslp) + arg_len;
393         }
394
395         free(lens);
396         clp->lines = cpu_to_le16(client->argc);
397         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
398         free(pdu);
399         return ret;
400 }
401
402 int fio_clients_connect(void)
403 {
404         struct fio_client *client;
405         struct flist_head *entry, *tmp;
406         int ret;
407
408 #ifdef WIN32
409         WSADATA wsd;
410         WSAStartup(MAKEWORD(2,2), &wsd);
411 #endif
412
413         dprint(FD_NET, "client: connect all\n");
414
415         client_signal_handler();
416
417         flist_for_each_safe(entry, tmp, &client_list) {
418                 client = flist_entry(entry, struct fio_client, list);
419
420                 ret = fio_client_connect(client);
421                 if (ret) {
422                         remove_client(client);
423                         continue;
424                 }
425
426                 probe_client(client);
427
428                 if (client->argc > 1)
429                         send_client_cmd_line(client);
430         }
431
432         return !nr_clients;
433 }
434
435 /*
436  * Send file contents to server backend. We could use sendfile(), but to remain
437  * more portable lets just read/write the darn thing.
438  */
439 static int fio_client_send_ini(struct fio_client *client, const char *filename)
440 {
441         struct stat sb;
442         char *p, *buf;
443         off_t len;
444         int fd, ret;
445
446         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
447
448         fd = open(filename, O_RDONLY);
449         if (fd < 0) {
450                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
451                 return 1;
452         }
453
454         if (fstat(fd, &sb) < 0) {
455                 log_err("fio: job file stat: %s\n", strerror(errno));
456                 close(fd);
457                 return 1;
458         }
459
460         buf = malloc(sb.st_size);
461
462         len = sb.st_size;
463         p = buf;
464         do {
465                 ret = read(fd, p, len);
466                 if (ret > 0) {
467                         len -= ret;
468                         if (!len)
469                                 break;
470                         p += ret;
471                         continue;
472                 } else if (!ret)
473                         break;
474                 else if (errno == EAGAIN || errno == EINTR)
475                         continue;
476         } while (1);
477
478         if (len) {
479                 log_err("fio: failed reading job file %s\n", filename);
480                 close(fd);
481                 free(buf);
482                 return 1;
483         }
484
485         client->sent_job = 1;
486         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size, 0);
487         free(buf);
488         close(fd);
489         return ret;
490 }
491
492 int fio_clients_send_ini(const char *filename)
493 {
494         struct fio_client *client;
495         struct flist_head *entry, *tmp;
496
497         flist_for_each_safe(entry, tmp, &client_list) {
498                 client = flist_entry(entry, struct fio_client, list);
499
500                 if (fio_client_send_ini(client, filename))
501                         remove_client(client);
502
503                 client->sent_job = 1;
504         }
505
506         return !nr_clients;
507 }
508
509 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
510 {
511         dst->max_val    = le64_to_cpu(src->max_val);
512         dst->min_val    = le64_to_cpu(src->min_val);
513         dst->samples    = le64_to_cpu(src->samples);
514
515         /*
516          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
517          */
518         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
519         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
520 }
521
522 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
523 {
524         int i, j;
525
526         dst->error      = le32_to_cpu(src->error);
527         dst->groupid    = le32_to_cpu(src->groupid);
528         dst->pid        = le32_to_cpu(src->pid);
529         dst->members    = le32_to_cpu(src->members);
530
531         for (i = 0; i < 2; i++) {
532                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
533                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
534                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
535                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
536         }
537
538         dst->usr_time           = le64_to_cpu(src->usr_time);
539         dst->sys_time           = le64_to_cpu(src->sys_time);
540         dst->ctx                = le64_to_cpu(src->ctx);
541         dst->minf               = le64_to_cpu(src->minf);
542         dst->majf               = le64_to_cpu(src->majf);
543         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
544
545         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
546                 fio_fp64_t *fps = &src->percentile_list[i];
547                 fio_fp64_t *fpd = &dst->percentile_list[i];
548
549                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
550         }
551
552         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
553                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
554                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
555                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
556         }
557
558         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
559                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
560                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
561         }
562
563         for (i = 0; i < 2; i++)
564                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
565                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
566
567         for (i = 0; i < 3; i++) {
568                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
569                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
570         }
571
572         dst->total_submit       = le64_to_cpu(src->total_submit);
573         dst->total_complete     = le64_to_cpu(src->total_complete);
574
575         for (i = 0; i < 2; i++) {
576                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
577                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
578         }
579
580         dst->total_run_time     = le64_to_cpu(src->total_run_time);
581         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
582         dst->total_err_count    = le64_to_cpu(src->total_err_count);
583         dst->first_error        = le32_to_cpu(src->first_error);
584         dst->kb_base            = le32_to_cpu(src->kb_base);
585 }
586
587 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
588 {
589         int i;
590
591         for (i = 0; i < 2; i++) {
592                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
593                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
594                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
595                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
596                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
597                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
598         }
599
600         dst->kb_base    = le32_to_cpu(src->kb_base);
601         dst->groupid    = le32_to_cpu(src->groupid);
602 }
603
604 static void handle_ts(struct fio_net_cmd *cmd)
605 {
606         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
607
608         convert_ts(&p->ts, &p->ts);
609         convert_gs(&p->rs, &p->rs);
610
611         show_thread_status(&p->ts, &p->rs);
612
613         if (sum_stat_clients == 1)
614                 return;
615
616         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
617         sum_group_stats(&client_gs, &p->rs);
618
619         client_ts.members++;
620         client_ts.groupid = p->ts.groupid;
621
622         if (++sum_stat_nr == sum_stat_clients) {
623                 strcpy(client_ts.name, "All clients");
624                 show_thread_status(&client_ts, &client_gs);
625         }
626 }
627
628 static void handle_gs(struct fio_net_cmd *cmd)
629 {
630         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
631
632         convert_gs(gs, gs);
633         show_group_stats(gs);
634 }
635
636 static void convert_agg(struct disk_util_agg *agg)
637 {
638         int i;
639
640         for (i = 0; i < 2; i++) {
641                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
642                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
643                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
644                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
645         }
646
647         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
648         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
649         agg->slavecount         = le32_to_cpu(agg->slavecount);
650         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
651 }
652
653 static void convert_dus(struct disk_util_stat *dus)
654 {
655         int i;
656
657         for (i = 0; i < 2; i++) {
658                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
659                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
660                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
661                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
662         }
663
664         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
665         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
666         dus->msec               = le64_to_cpu(dus->msec);
667 }
668
669 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
670 {
671         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
672
673         convert_dus(&du->dus);
674         convert_agg(&du->agg);
675
676         if (!client->disk_stats_shown) {
677                 client->disk_stats_shown = 1;
678                 log_info("\nDisk stats (read/write):\n");
679         }
680
681         print_disk_util(&du->dus, &du->agg, terse_output);
682 }
683
684 static void convert_jobs_eta(struct jobs_eta *je)
685 {
686         int i;
687
688         je->nr_running          = le32_to_cpu(je->nr_running);
689         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
690         je->nr_pending          = le32_to_cpu(je->nr_pending);
691         je->files_open          = le32_to_cpu(je->files_open);
692         je->m_rate              = le32_to_cpu(je->m_rate);
693         je->t_rate              = le32_to_cpu(je->t_rate);
694         je->m_iops              = le32_to_cpu(je->m_iops);
695         je->t_iops              = le32_to_cpu(je->t_iops);
696
697         for (i = 0; i < 2; i++) {
698                 je->rate[i]     = le32_to_cpu(je->rate[i]);
699                 je->iops[i]     = le32_to_cpu(je->iops[i]);
700         }
701
702         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
703         je->eta_sec             = le64_to_cpu(je->eta_sec);
704 }
705
706 static void sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
707 {
708         int i;
709
710         dst->nr_running         += je->nr_running;
711         dst->nr_ramp            += je->nr_ramp;
712         dst->nr_pending         += je->nr_pending;
713         dst->files_open         += je->files_open;
714         dst->m_rate             += je->m_rate;
715         dst->t_rate             += je->t_rate;
716         dst->m_iops             += je->m_iops;
717         dst->t_iops             += je->t_iops;
718
719         for (i = 0; i < 2; i++) {
720                 dst->rate[i]    += je->rate[i];
721                 dst->iops[i]    += je->iops[i];
722         }
723
724         dst->elapsed_sec        += je->elapsed_sec;
725
726         if (je->eta_sec > dst->eta_sec)
727                 dst->eta_sec = je->eta_sec;
728 }
729
730 static void dec_jobs_eta(struct client_eta *eta)
731 {
732         if (!--eta->pending) {
733                 display_thread_status(&eta->eta);
734                 free(eta);
735         }
736 }
737
738 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
739 {
740         struct fio_net_int_cmd *icmd = NULL;
741         struct flist_head *entry;
742
743         flist_for_each(entry, &client->cmd_list) {
744                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
745
746                 if (cmd->tag == (uintptr_t) icmd)
747                         break;
748
749                 icmd = NULL;
750         }
751
752         if (!icmd) {
753                 log_err("fio: client: unable to find matching tag\n");
754                 return;
755         }
756
757         flist_del(&icmd->list);
758         cmd->tag = icmd->saved_tag;
759         free(icmd);
760 }
761
762 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
763 {
764         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
765         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
766
767         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
768
769         assert(client->eta_in_flight == eta);
770
771         client->eta_in_flight = NULL;
772         flist_del_init(&client->eta_list);
773
774         convert_jobs_eta(je);
775         sum_jobs_eta(&eta->eta, je);
776         dec_jobs_eta(eta);
777 }
778
779 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
780 {
781         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
782         const char *os, *arch;
783         char bit[16];
784
785         os = fio_get_os_string(probe->os);
786         if (!os)
787                 os = "unknown";
788
789         arch = fio_get_arch_string(probe->arch);
790         if (!arch)
791                 os = "unknown";
792
793         sprintf(bit, "%d-bit", probe->bpp * 8);
794
795         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
796                 probe->hostname, probe->bigendian, bit, os, arch,
797                 probe->fio_major, probe->fio_minor, probe->fio_patch);
798
799         if (!client->name)
800                 client->name = strdup((char *) probe->hostname);
801 }
802
803 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
804 {
805         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
806
807         client->state = Client_started;
808         client->jobs = le32_to_cpu(pdu->jobs);
809 }
810
811 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
812 {
813         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
814
815         client->state = Client_stopped;
816         client->error = le32_to_cpu(pdu->error);
817
818         if (client->error)
819                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
820 }
821
822 static int handle_client(struct fio_client *client)
823 {
824         struct fio_net_cmd *cmd;
825
826         dprint(FD_NET, "client: handle %s\n", client->hostname);
827
828         cmd = fio_net_recv_cmd(client->fd);
829         if (!cmd)
830                 return 0;
831
832         dprint(FD_NET, "client: got cmd op %s from %s\n",
833                                 fio_server_op(cmd->opcode), client->hostname);
834
835         switch (cmd->opcode) {
836         case FIO_NET_CMD_QUIT:
837                 remove_client(client);
838                 free(cmd);
839                 break;
840         case FIO_NET_CMD_TEXT: {
841                 const char *buf = (const char *) cmd->payload;
842                 const char *name;
843                 int fio_unused ret;
844
845                 name = client->name ? client->name : client->hostname;
846
847                 if (!client->skip_newline)
848                         fprintf(f_out, "<%s> ", name);
849                 ret = fwrite(buf, cmd->pdu_len, 1, f_out);
850                 fflush(f_out);
851                 client->skip_newline = strchr(buf, '\n') == NULL;
852                 free(cmd);
853                 break;
854                 }
855         case FIO_NET_CMD_DU:
856                 handle_du(client, cmd);
857                 free(cmd);
858                 break;
859         case FIO_NET_CMD_TS:
860                 handle_ts(cmd);
861                 free(cmd);
862                 break;
863         case FIO_NET_CMD_GS:
864                 handle_gs(cmd);
865                 free(cmd);
866                 break;
867         case FIO_NET_CMD_ETA:
868                 remove_reply_cmd(client, cmd);
869                 handle_eta(client, cmd);
870                 free(cmd);
871                 break;
872         case FIO_NET_CMD_PROBE:
873                 remove_reply_cmd(client, cmd);
874                 handle_probe(client, cmd);
875                 free(cmd);
876                 break;
877         case FIO_NET_CMD_RUN:
878                 client->state = Client_running;
879                 free(cmd);
880                 break;
881         case FIO_NET_CMD_START:
882                 handle_start(client, cmd);
883                 free(cmd);
884                 break;
885         case FIO_NET_CMD_STOP:
886                 handle_stop(client, cmd);
887                 free(cmd);
888                 break;
889         default:
890                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
891                 free(cmd);
892                 break;
893         }
894
895         return 1;
896 }
897
898 static void request_client_etas(void)
899 {
900         struct fio_client *client;
901         struct flist_head *entry;
902         struct client_eta *eta;
903         int skipped = 0;
904
905         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
906
907         eta = malloc(sizeof(*eta));
908         memset(&eta->eta, 0, sizeof(eta->eta));
909         eta->pending = nr_clients;
910
911         flist_for_each(entry, &client_list) {
912                 client = flist_entry(entry, struct fio_client, list);
913
914                 if (!flist_empty(&client->eta_list)) {
915                         skipped++;
916                         continue;
917                 }
918                 if (client->state != Client_running)
919                         continue;
920
921                 assert(!client->eta_in_flight);
922                 flist_add_tail(&client->eta_list, &eta_list);
923                 client->eta_in_flight = eta;
924                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
925                                         (uintptr_t) eta, &client->cmd_list);
926         }
927
928         while (skipped--)
929                 dec_jobs_eta(eta);
930
931         dprint(FD_NET, "client: requested eta tag %p\n", eta);
932 }
933
934 static int client_check_cmd_timeout(struct fio_client *client,
935                                     struct timeval *now)
936 {
937         struct fio_net_int_cmd *cmd;
938         struct flist_head *entry, *tmp;
939         int ret = 0;
940
941         flist_for_each_safe(entry, tmp, &client->cmd_list) {
942                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
943
944                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
945                         continue;
946
947                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
948                                                 fio_server_op(cmd->cmd.opcode));
949                 flist_del(&cmd->list);
950                 free(cmd);
951                 ret = 1;
952         }
953
954         return flist_empty(&client->cmd_list) && ret;
955 }
956
957 static int fio_client_timed_out(void)
958 {
959         struct fio_client *client;
960         struct flist_head *entry, *tmp;
961         struct timeval tv;
962         int ret = 0;
963
964         gettimeofday(&tv, NULL);
965
966         flist_for_each_safe(entry, tmp, &client_list) {
967                 client = flist_entry(entry, struct fio_client, list);
968
969                 if (flist_empty(&client->cmd_list))
970                         continue;
971
972                 if (!client_check_cmd_timeout(client, &tv))
973                         continue;
974
975                 log_err("fio: client %s timed out\n", client->hostname);
976                 remove_client(client);
977                 ret = 1;
978         }
979
980         return ret;
981 }
982
983 int fio_handle_clients(void)
984 {
985         struct pollfd *pfds;
986         int i, ret = 0, retval = 0;
987
988         gettimeofday(&eta_tv, NULL);
989
990         pfds = malloc(nr_clients * sizeof(struct pollfd));
991
992         sum_stat_clients = nr_clients;
993         init_thread_stat(&client_ts);
994         init_group_run_stat(&client_gs);
995
996         while (!exit_backend && nr_clients) {
997                 struct flist_head *entry, *tmp;
998                 struct fio_client *client;
999
1000                 i = 0;
1001                 flist_for_each_safe(entry, tmp, &client_list) {
1002                         client = flist_entry(entry, struct fio_client, list);
1003
1004                         if (!client->sent_job &&
1005                             flist_empty(&client->cmd_list)) {
1006                                 remove_client(client);
1007                                 continue;
1008                         }
1009
1010                         pfds[i].fd = client->fd;
1011                         pfds[i].events = POLLIN;
1012                         i++;
1013                 }
1014
1015                 if (!nr_clients)
1016                         break;
1017
1018                 assert(i == nr_clients);
1019
1020                 do {
1021                         struct timeval tv;
1022
1023                         gettimeofday(&tv, NULL);
1024                         if (mtime_since(&eta_tv, &tv) >= 900) {
1025                                 request_client_etas();
1026                                 memcpy(&eta_tv, &tv, sizeof(tv));
1027
1028                                 if (fio_client_timed_out())
1029                                         break;
1030                         }
1031
1032                         ret = poll(pfds, nr_clients, 100);
1033                         if (ret < 0) {
1034                                 if (errno == EINTR)
1035                                         continue;
1036                                 log_err("fio: poll clients: %s\n", strerror(errno));
1037                                 break;
1038                         } else if (!ret)
1039                                 continue;
1040                 } while (ret <= 0);
1041
1042                 for (i = 0; i < nr_clients; i++) {
1043                         if (!(pfds[i].revents & POLLIN))
1044                                 continue;
1045
1046                         client = find_client_by_fd(pfds[i].fd);
1047                         if (!client) {
1048                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1049                                 continue;
1050                         }
1051                         if (!handle_client(client)) {
1052                                 log_info("client: host=%s disconnected\n",
1053                                                 client->hostname);
1054                                 remove_client(client);
1055                                 retval = 1;
1056                         } else if (client->error)
1057                                 retval = 1;
1058                 }
1059         }
1060
1061         free(pfds);
1062         return retval;
1063 }