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