Remember to close sockets on error
[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 fio_client {
24         struct flist_head list;
25         struct flist_head fd_hash_list;
26         struct flist_head name_hash_list;
27         struct sockaddr_in addr;
28         struct sockaddr_un addr_un;
29         char *hostname;
30         int fd;
31
32         int state;
33         int skip_newline;
34         int is_sock;
35
36         uint16_t argc;
37         char **argv;
38 };
39
40 enum {
41         Client_created          = 0,
42         Client_connected        = 1,
43         Client_started          = 2,
44         Client_stopped          = 3,
45         Client_exited           = 4,
46 };
47
48 static FLIST_HEAD(client_list);
49
50 #define FIO_CLIENT_HASH_BITS    7
51 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
52 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
53 static struct flist_head client_fd_hash[FIO_CLIENT_HASH_SZ];
54 static struct flist_head client_name_hash[FIO_CLIENT_HASH_SZ];
55
56 static int handle_client(struct fio_client *client);
57
58 static void fio_client_add_fd_hash(struct fio_client *client)
59 {
60         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
61
62         bucket &= FIO_CLIENT_HASH_MASK;
63         flist_add(&client->fd_hash_list, &client_fd_hash[bucket]);
64 }
65
66 static void fio_client_remove_fd_hash(struct fio_client *client)
67 {
68         if (!flist_empty(&client->fd_hash_list))
69                 flist_del_init(&client->fd_hash_list);
70 }
71
72 static void fio_client_add_name_hash(struct fio_client *client)
73 {
74         int bucket = jhash(client->hostname, strlen(client->hostname), 0);
75
76         bucket &= FIO_CLIENT_HASH_MASK;
77         flist_add(&client->name_hash_list, &client_name_hash[bucket]);
78 }
79
80 static void fio_client_remove_name_hash(struct fio_client *client)
81 {
82         if (!flist_empty(&client->name_hash_list))
83                 flist_del_init(&client->name_hash_list);
84 }
85
86 static void fio_init fio_client_hash_init(void)
87 {
88         int i;
89
90         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++) {
91                 INIT_FLIST_HEAD(&client_fd_hash[i]);
92                 INIT_FLIST_HEAD(&client_name_hash[i]);
93         }
94 }
95
96 static struct fio_client *find_client_by_fd(int fd)
97 {
98         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
99         struct fio_client *client;
100         struct flist_head *entry;
101
102         flist_for_each(entry, &client_fd_hash[bucket]) {
103                 client = flist_entry(entry, struct fio_client, fd_hash_list);
104
105                 if (client->fd == fd)
106                         return client;
107         }
108
109         return NULL;
110 }
111
112 static struct fio_client *find_client_by_name(const char *name)
113 {
114         int bucket = jhash(name, strlen(name), 0) & FIO_CLIENT_HASH_BITS;
115         struct fio_client *client;
116         struct flist_head *entry;
117
118         flist_for_each(entry, &client_name_hash[bucket]) {
119                 client = flist_entry(entry, struct fio_client, name_hash_list);
120
121                 if (!strcmp(name, client->hostname))
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_fd_hash(client);
134         fio_client_remove_name_hash(client);
135
136         free(client->hostname);
137         if (client->argv)
138                 free(client->argv);
139
140         free(client);
141         nr_clients--;
142 }
143
144 static int __fio_client_add_cmd_option(struct fio_client *client,
145                                        const char *opt)
146 {
147         int index;
148
149         if (client->argc == FIO_NET_CMD_JOBLINE_ARGV) {
150                 log_err("fio: max cmd line number reached.\n");
151                 log_err("fio: cmd line <%s> has been ignored.\n", opt);
152                 return 1;
153         }
154
155         index = client->argc++;
156         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
157         client->argv[index] = strdup(opt);
158         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
159         return 0;
160 }
161
162 int fio_client_add_cmd_option(const char *hostname, const char *opt)
163 {
164         struct fio_client *client;
165
166         if (!hostname || !opt)
167                 return 0;
168
169         client = find_client_by_name(hostname);
170         if (!client) {
171                 log_err("fio: unknown client %s\n", hostname);
172                 return 1;
173         }
174
175         return __fio_client_add_cmd_option(client, opt);
176 }
177
178 void fio_client_add(const char *hostname)
179 {
180         struct fio_client *client;
181
182         dprint(FD_NET, "client: added  <%s>\n", hostname);
183         client = malloc(sizeof(*client));
184         memset(client, 0, sizeof(*client));
185
186         INIT_FLIST_HEAD(&client->list);
187         INIT_FLIST_HEAD(&client->fd_hash_list);
188         INIT_FLIST_HEAD(&client->name_hash_list);
189
190         if (!strncmp(hostname, "sock:", 5)) {
191                 client->hostname = strdup(hostname + 5);
192                 client->is_sock = 1;
193         } else
194                 client->hostname = strdup(hostname);
195
196         client->fd = -1;
197
198         fio_client_add_name_hash(client);
199
200         __fio_client_add_cmd_option(client, "fio");
201
202         flist_add(&client->list, &client_list);
203         nr_clients++;
204 }
205
206 static int fio_client_connect_ip(struct fio_client *client)
207 {
208         int fd;
209
210         client->addr.sin_family = AF_INET;
211         client->addr.sin_port = htons(fio_net_port);
212
213         if (inet_aton(client->hostname, &client->addr.sin_addr) != 1) {
214                 struct hostent *hent;
215
216                 hent = gethostbyname(client->hostname);
217                 if (!hent) {
218                         log_err("fio: gethostbyname: %s\n", strerror(errno));
219                         log_err("fio: failed looking up hostname %s\n",
220                                         client->hostname);
221                         return -1;
222                 }
223
224                 memcpy(&client->addr.sin_addr, hent->h_addr, 4);
225         }
226
227         fd = socket(AF_INET, SOCK_STREAM, 0);
228         if (fd < 0) {
229                 log_err("fio: socket: %s\n", strerror(errno));
230                 return -1;
231         }
232
233         if (connect(fd, (struct sockaddr *) &client->addr, sizeof(client->addr)) < 0) {
234                 log_err("fio: connect: %s\n", strerror(errno));
235                 log_err("fio: failed to connect to %s\n", client->hostname);
236                 close(fd);
237                 return -1;
238         }
239
240         return fd;
241 }
242
243 static int fio_client_connect_sock(struct fio_client *client)
244 {
245         struct sockaddr_un *addr = &client->addr_un;
246         fio_socklen_t len;
247         int fd;
248
249         memset(addr, 0, sizeof(*addr));
250         addr->sun_family = AF_UNIX;
251         strcpy(addr->sun_path, client->hostname);
252
253         fd = socket(AF_UNIX, SOCK_STREAM, 0);
254         if (fd < 0) {
255                 log_err("fio: socket: %s\n", strerror(errno));
256                 return -1;
257         }
258
259         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
260         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
261                 log_err("fio: connect; %s\n", strerror(errno));
262                 close(fd);
263                 return -1;
264         }
265
266         return fd;
267 }
268
269 static int fio_client_connect(struct fio_client *client)
270 {
271         int fd;
272
273         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
274
275         memset(&client->addr, 0, sizeof(client->addr));
276
277         if (client->is_sock)
278                 fd = fio_client_connect_sock(client);
279         else
280                 fd = fio_client_connect_ip(client);
281
282         if (fd < 0)
283                 return 1;
284
285         client->fd = fd;
286         fio_client_add_fd_hash(client);
287         client->state = Client_connected;
288         return 0;
289 }
290
291 void fio_clients_terminate(void)
292 {
293         struct flist_head *entry;
294         struct fio_client *client;
295
296         dprint(FD_NET, "client: terminate clients\n");
297
298         flist_for_each(entry, &client_list) {
299                 client = flist_entry(entry, struct fio_client, list);
300
301                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0);
302         }
303 }
304
305 static void sig_int(int sig)
306 {
307         dprint(FD_NET, "client: got sign %d\n", sig);
308         fio_clients_terminate();
309 }
310
311 static void client_signal_handler(void)
312 {
313         struct sigaction act;
314
315         memset(&act, 0, sizeof(act));
316         act.sa_handler = sig_int;
317         act.sa_flags = SA_RESTART;
318         sigaction(SIGINT, &act, NULL);
319
320         memset(&act, 0, sizeof(act));
321         act.sa_handler = sig_int;
322         act.sa_flags = SA_RESTART;
323         sigaction(SIGTERM, &act, NULL);
324 }
325
326 static void probe_client(struct fio_client *client)
327 {
328         dprint(FD_NET, "client: send probe\n");
329
330         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0);
331         handle_client(client);
332 }
333
334 static int send_client_cmd_line(struct fio_client *client)
335 {
336         struct cmd_line_pdu *pdu;
337         int i, ret;
338
339         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
340
341         pdu = malloc(sizeof(*pdu));
342         for (i = 0; i < client->argc; i++)
343                 strcpy((char *) pdu->argv[i], client->argv[i]);
344
345         pdu->argc = cpu_to_le16(client->argc);
346         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, sizeof(*pdu));
347         free(pdu);
348         return ret;
349 }
350
351 int fio_clients_connect(void)
352 {
353         struct fio_client *client;
354         struct flist_head *entry, *tmp;
355         int ret;
356
357         dprint(FD_NET, "client: connect all\n");
358
359         client_signal_handler();
360
361         flist_for_each_safe(entry, tmp, &client_list) {
362                 client = flist_entry(entry, struct fio_client, list);
363
364                 ret = fio_client_connect(client);
365                 if (ret) {
366                         remove_client(client);
367                         continue;
368                 }
369
370                 probe_client(client);
371
372                 if (client->argc > 1)
373                         send_client_cmd_line(client);
374         }
375
376         return !nr_clients;
377 }
378
379 /*
380  * Send file contents to server backend. We could use sendfile(), but to remain
381  * more portable lets just read/write the darn thing.
382  */
383 static int fio_client_send_ini(struct fio_client *client, const char *filename)
384 {
385         struct stat sb;
386         char *p, *buf;
387         off_t len;
388         int fd, ret;
389
390         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
391
392         fd = open(filename, O_RDONLY);
393         if (fd < 0) {
394                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
395                 return 1;
396         }
397
398         if (fstat(fd, &sb) < 0) {
399                 log_err("fio: job file stat: %s\n", strerror(errno));
400                 close(fd);
401                 return 1;
402         }
403
404         buf = malloc(sb.st_size);
405
406         len = sb.st_size;
407         p = buf;
408         do {
409                 ret = read(fd, p, len);
410                 if (ret > 0) {
411                         len -= ret;
412                         if (!len)
413                                 break;
414                         p += ret;
415                         continue;
416                 } else if (!ret)
417                         break;
418                 else if (errno == EAGAIN || errno == EINTR)
419                         continue;
420         } while (1);
421
422         if (len) {
423                 log_err("fio: failed reading job file %s\n", filename);
424                 close(fd);
425                 return 1;
426         }
427
428         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size);
429         free(buf);
430         close(fd);
431         return ret;
432 }
433
434 int fio_clients_send_ini(const char *filename)
435 {
436         struct fio_client *client;
437         struct flist_head *entry, *tmp;
438
439         flist_for_each_safe(entry, tmp, &client_list) {
440                 client = flist_entry(entry, struct fio_client, list);
441
442                 if (fio_client_send_ini(client, filename))
443                         remove_client(client);
444         }
445
446         return !nr_clients;
447 }
448
449 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
450 {
451         dst->max_val    = le64_to_cpu(src->max_val);
452         dst->min_val    = le64_to_cpu(src->min_val);
453         dst->samples    = le64_to_cpu(src->samples);
454
455         /*
456          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
457          */
458         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
459         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
460 }
461
462 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
463 {
464         int i, j;
465
466         dst->error      = le32_to_cpu(src->error);
467         dst->groupid    = le32_to_cpu(src->groupid);
468         dst->pid        = le32_to_cpu(src->pid);
469         dst->members    = le32_to_cpu(src->members);
470
471         for (i = 0; i < 2; i++) {
472                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
473                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
474                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
475                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
476         }
477
478         dst->usr_time           = le64_to_cpu(src->usr_time);
479         dst->sys_time           = le64_to_cpu(src->sys_time);
480         dst->ctx                = le64_to_cpu(src->ctx);
481         dst->minf               = le64_to_cpu(src->minf);
482         dst->majf               = le64_to_cpu(src->majf);
483         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
484
485         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
486                 fio_fp64_t *fps = &src->percentile_list[i];
487                 fio_fp64_t *fpd = &dst->percentile_list[i];
488
489                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
490         }
491
492         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
493                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
494                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
495                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
496         }
497
498         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
499                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
500                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
501         }
502
503         for (i = 0; i < 2; i++)
504                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
505                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
506
507         for (i = 0; i < 3; i++) {
508                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
509                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
510         }
511
512         dst->total_submit       = le64_to_cpu(src->total_submit);
513         dst->total_complete     = le64_to_cpu(src->total_complete);
514
515         for (i = 0; i < 2; i++) {
516                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
517                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
518         }
519
520         dst->total_run_time     = le64_to_cpu(src->total_run_time);
521         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
522         dst->total_err_count    = le64_to_cpu(src->total_err_count);
523         dst->first_error        = le32_to_cpu(src->first_error);
524         dst->kb_base            = le32_to_cpu(src->kb_base);
525 }
526
527 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
528 {
529         int i;
530
531         for (i = 0; i < 2; i++) {
532                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
533                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
534                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
535                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
536                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
537                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
538         }
539
540         dst->kb_base    = le32_to_cpu(src->kb_base);
541         dst->groupid    = le32_to_cpu(src->groupid);
542 }
543
544 static void handle_ts(struct fio_net_cmd *cmd)
545 {
546         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
547
548         convert_ts(&p->ts, &p->ts);
549         convert_gs(&p->rs, &p->rs);
550
551         show_thread_status(&p->ts, &p->rs);
552 }
553
554 static void handle_gs(struct fio_net_cmd *cmd)
555 {
556         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
557
558         convert_gs(gs, gs);
559         show_group_stats(gs);
560 }
561
562 static void handle_eta(struct fio_net_cmd *cmd)
563 {
564         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
565         int i;
566
567         je->nr_running          = le32_to_cpu(je->nr_running);
568         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
569         je->nr_pending          = le32_to_cpu(je->nr_pending);
570         je->files_open          = le32_to_cpu(je->files_open);
571         je->m_rate              = le32_to_cpu(je->m_rate);
572         je->t_rate              = le32_to_cpu(je->t_rate);
573         je->m_iops              = le32_to_cpu(je->m_iops);
574         je->t_iops              = le32_to_cpu(je->t_iops);
575
576         for (i = 0; i < 2; i++) {
577                 je->rate[i]     = le32_to_cpu(je->rate[i]);
578                 je->iops[i]     = le32_to_cpu(je->iops[i]);
579         }
580
581         je->elapsed_sec         = le32_to_cpu(je->nr_running);
582         je->eta_sec             = le64_to_cpu(je->eta_sec);
583
584         display_thread_status(je);
585 }
586
587 static void handle_probe(struct fio_net_cmd *cmd)
588 {
589         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
590
591         log_info("Probe: hostname=%s, be=%u, fio ver %u.%u.%u\n",
592                 probe->hostname, probe->bigendian, probe->fio_major,
593                 probe->fio_minor, probe->fio_patch);
594 }
595
596 static int handle_client(struct fio_client *client)
597 {
598         struct fio_net_cmd *cmd;
599
600         dprint(FD_NET, "client: handle %s\n", client->hostname);
601
602         cmd = fio_net_recv_cmd(client->fd);
603         if (!cmd)
604                 return 0;
605
606         dprint(FD_NET, "client: got cmd op %d from %s\n",
607                                         cmd->opcode, client->hostname);
608
609         switch (cmd->opcode) {
610         case FIO_NET_CMD_QUIT:
611                 remove_client(client);
612                 free(cmd);
613                 break;
614         case FIO_NET_CMD_TEXT: {
615                 const char *buf = (const char *) cmd->payload;
616                 int fio_unused ret;
617
618                 if (!client->skip_newline)
619                         fprintf(f_out, "<%s> ", client->hostname);
620                 ret = fwrite(buf, cmd->pdu_len, 1, f_out);
621                 fflush(f_out);
622                 client->skip_newline = strchr(buf, '\n') == NULL;
623                 free(cmd);
624                 break;
625                 }
626         case FIO_NET_CMD_TS:
627                 handle_ts(cmd);
628                 free(cmd);
629                 break;
630         case FIO_NET_CMD_GS:
631                 handle_gs(cmd);
632                 free(cmd);
633                 break;
634         case FIO_NET_CMD_ETA:
635                 handle_eta(cmd);
636                 free(cmd);
637                 break;
638         case FIO_NET_CMD_PROBE:
639                 handle_probe(cmd);
640                 free(cmd);
641                 break;
642         case FIO_NET_CMD_START:
643                 client->state = Client_started;
644                 free(cmd);
645                 break;
646         case FIO_NET_CMD_STOP:
647                 client->state = Client_stopped;
648                 free(cmd);
649                 break;
650         default:
651                 log_err("fio: unknown client op: %d\n", cmd->opcode);
652                 free(cmd);
653                 break;
654         }
655
656         return 1;
657 }
658
659 int fio_handle_clients(void)
660 {
661         struct fio_client *client;
662         struct flist_head *entry;
663         struct pollfd *pfds;
664         int i, ret = 0;
665
666         pfds = malloc(nr_clients * sizeof(struct pollfd));
667
668         while (!exit_backend && nr_clients) {
669                 i = 0;
670                 flist_for_each(entry, &client_list) {
671                         client = flist_entry(entry, struct fio_client, list);
672
673                         pfds[i].fd = client->fd;
674                         pfds[i].events = POLLIN;
675                         i++;
676                 }
677
678                 assert(i == nr_clients);
679
680                 do {
681                         ret = poll(pfds, nr_clients, 100);
682                         if (ret < 0) {
683                                 if (errno == EINTR)
684                                         continue;
685                                 log_err("fio: poll clients: %s\n", strerror(errno));
686                                 break;
687                         } else if (!ret)
688                                 continue;
689                 } while (ret <= 0);
690
691                 for (i = 0; i < nr_clients; i++) {
692                         if (!(pfds[i].revents & POLLIN))
693                                 continue;
694
695                         client = find_client_by_fd(pfds[i].fd);
696                         if (!client) {
697                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
698                                 continue;
699                         }
700                         if (!handle_client(client)) {
701                                 log_info("client: host=%s disconnected\n",
702                                                 client->hostname);
703                                 remove_client(client);
704                         }
705                 }
706         }
707
708         free(pfds);
709         return 0;
710 }