server: add a few more debug logging statements
[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/mman.h>
12 #include <netinet/in.h>
13 #include <arpa/inet.h>
14 #include <netdb.h>
15
16 #include "fio.h"
17 #include "server.h"
18 #include "crc/crc32.h"
19 #include "flist.h"
20
21 struct fio_client {
22         struct flist_head list;
23         struct sockaddr_in addr;
24         char *hostname;
25         int fd;
26
27         int state;
28         int skip_newline;
29
30         uint16_t argc;
31         char **argv;
32 };
33
34 enum {
35         Client_created          = 0,
36         Client_connected        = 1,
37         Client_started          = 2,
38         Client_stopped          = 3,
39         Client_exited           = 4,
40 };
41
42 static FLIST_HEAD(client_list);
43
44 static int handle_client(struct fio_client *client, int one);
45
46 static struct fio_client *find_client_by_fd(int fd)
47 {
48         struct fio_client *client;
49         struct flist_head *entry;
50
51         flist_for_each(entry, &client_list) {
52                 client = flist_entry(entry, struct fio_client, list);
53
54                 if (client->fd == fd)
55                         return client;
56         }
57
58         return NULL;
59 }
60
61 static struct fio_client *find_client_by_name(const char *name)
62 {
63         struct fio_client *client;
64         struct flist_head *entry;
65
66         flist_for_each(entry, &client_list) {
67                 client = flist_entry(entry, struct fio_client, list);
68
69                 if (!strcmp(name, client->hostname))
70                         return client;
71         }
72
73         return NULL;
74 }
75
76 static void remove_client(struct fio_client *client)
77 {
78         dprint(FD_NET, "removed client <%s>\n", client->hostname);
79         flist_del(&client->list);
80         nr_clients--;
81
82         free(client->hostname);
83         if (client->argv)
84                 free(client->argv);
85
86         free(client);
87 }
88
89 static void __fio_client_add_cmd_option(struct fio_client *client,
90                                         const char *opt)
91 {
92         client->argc++;
93         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
94         client->argv[client->argc - 1] = strdup(opt);
95 }
96
97 void fio_client_add_cmd_option(const char *hostname, const char *opt)
98 {
99         struct fio_client *client;
100
101         if (!hostname || !opt)
102                 return;
103
104         client = find_client_by_name(hostname);
105         if (!client) {
106                 log_err("fio: unknown client %s\n", hostname);
107                 return;
108         }
109
110         __fio_client_add_cmd_option(client, opt);
111 }
112
113 void fio_client_add(const char *hostname)
114 {
115         struct fio_client *client;
116
117         dprint(FD_NET, "added client <%s>\n", hostname);
118         client = malloc(sizeof(*client));
119         memset(client, 0, sizeof(*client));
120
121         client->hostname = strdup(hostname);
122         client->fd = -1;
123
124         __fio_client_add_cmd_option(client, "fio");
125
126         flist_add(&client->list, &client_list);
127         nr_clients++;
128 }
129
130 static int fio_client_connect(struct fio_client *client)
131 {
132         int fd;
133
134         dprint(FD_NET, "connect to host %s\n", client->hostname);
135
136         memset(&client->addr, 0, sizeof(client->addr));
137         client->addr.sin_family = AF_INET;
138         client->addr.sin_port = htons(fio_net_port);
139
140         if (inet_aton(client->hostname, &client->addr.sin_addr) != 1) {
141                 struct hostent *hent;
142
143                 hent = gethostbyname(client->hostname);
144                 if (!hent) {
145                         log_err("fio: gethostbyname: %s\n", strerror(errno));
146                         return 1;
147                 }
148
149                 memcpy(&client->addr.sin_addr, hent->h_addr, 4);
150         }
151
152         fd = socket(AF_INET, SOCK_STREAM, 0);
153         if (fd < 0) {
154                 log_err("fio: socket: %s\n", strerror(errno));
155                 return 1;
156         }
157
158         if (connect(fd, (struct sockaddr *) &client->addr, sizeof(client->addr)) < 0) {
159                 log_err("fio: connect: %s\n", strerror(errno));
160                 log_err("fio: failed to connect to %s\n", client->hostname);
161                 return 1;
162         }
163
164         client->fd = fd;
165         client->state = Client_connected;
166         return 0;
167 }
168
169 void fio_clients_terminate(void)
170 {
171         struct flist_head *entry;
172         struct fio_client *client;
173
174         dprint(FD_NET, "client: terminate clients\n");
175
176         flist_for_each(entry, &client_list) {
177                 client = flist_entry(entry, struct fio_client, list);
178
179                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0);
180         }
181 }
182
183 static void sig_int(int sig)
184 {
185         dprint(FD_NET, "client: got sign %d\n", sig);
186         fio_clients_terminate();
187 }
188
189 static void client_signal_handler(void)
190 {
191         struct sigaction act;
192
193         memset(&act, 0, sizeof(act));
194         act.sa_handler = sig_int;
195         act.sa_flags = SA_RESTART;
196         sigaction(SIGINT, &act, NULL);
197
198         memset(&act, 0, sizeof(act));
199         act.sa_handler = sig_int;
200         act.sa_flags = SA_RESTART;
201         sigaction(SIGTERM, &act, NULL);
202 }
203
204 static void probe_client(struct fio_client *client)
205 {
206         dprint(FD_NET, "client: send probe\n");
207
208         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0);
209         handle_client(client, 1);
210 }
211
212 static int send_client_cmd_line(struct fio_client *client)
213 {
214         struct cmd_line_pdu *pdu;
215         int i, ret;
216
217         dprint(FD_NET, "client: send cmdline\n");
218
219         pdu = malloc(sizeof(*pdu));
220         for (i = 0; i < client->argc; i++)
221                 strcpy((char *) pdu->argv[i], client->argv[i]);
222
223         pdu->argc = cpu_to_le16(client->argc);
224         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, sizeof(*pdu));
225         free(pdu);
226         return ret;
227 }
228
229 int fio_clients_connect(void)
230 {
231         struct fio_client *client;
232         struct flist_head *entry, *tmp;
233         int ret;
234
235         dprint(FD_NET, "client: connect all\n");
236
237         client_signal_handler();
238
239         flist_for_each_safe(entry, tmp, &client_list) {
240                 client = flist_entry(entry, struct fio_client, list);
241
242                 ret = fio_client_connect(client);
243                 if (ret) {
244                         remove_client(client);
245                         continue;
246                 }
247
248                 probe_client(client);
249
250                 if (client->argc > 1)
251                         send_client_cmd_line(client);
252         }
253
254         return !nr_clients;
255 }
256
257 /*
258  * Send file contents to server backend. We could use sendfile(), but to remain
259  * more portable lets just read/write the darn thing.
260  */
261 static int fio_client_send_ini(struct fio_client *client, const char *filename)
262 {
263         struct stat sb;
264         char *p, *buf;
265         off_t len;
266         int fd, ret;
267
268         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
269
270         fd = open(filename, O_RDONLY);
271         if (fd < 0) {
272                 log_err("fio: job file open: %s\n", strerror(errno));
273                 return 1;
274         }
275
276         if (fstat(fd, &sb) < 0) {
277                 log_err("fio: job file stat: %s\n", strerror(errno));
278                 return 1;
279         }
280
281         buf = malloc(sb.st_size);
282
283         len = sb.st_size;
284         p = buf;
285         do {
286                 ret = read(fd, p, len);
287                 if (ret > 0) {
288                         len -= ret;
289                         if (!len)
290                                 break;
291                         p += ret;
292                         continue;
293                 } else if (!ret)
294                         break;
295                 else if (errno == EAGAIN || errno == EINTR)
296                         continue;
297         } while (1);
298
299         if (len) {
300                 log_err("fio: failed reading job file %s\n", filename);
301                 return 1;
302         }
303
304         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size);
305         free(buf);
306         return ret;
307 }
308
309 int fio_clients_send_ini(const char *filename)
310 {
311         struct fio_client *client;
312         struct flist_head *entry, *tmp;
313
314         flist_for_each_safe(entry, tmp, &client_list) {
315                 client = flist_entry(entry, struct fio_client, list);
316
317                 if (fio_client_send_ini(client, filename))
318                         remove_client(client);
319         }
320
321         return !nr_clients;
322 }
323
324 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
325 {
326         dst->max_val    = le64_to_cpu(src->max_val);
327         dst->min_val    = le64_to_cpu(src->min_val);
328         dst->samples    = le64_to_cpu(src->samples);
329         /* FIXME */
330         dst->mean       = __le64_to_cpu(src->mean);
331         dst->S          = __le64_to_cpu(src->S);
332 }
333
334 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
335 {
336         int i, j;
337
338         dst->error      = le32_to_cpu(src->error);
339         dst->groupid    = le32_to_cpu(src->groupid);
340         dst->pid        = le32_to_cpu(src->pid);
341         dst->members    = le32_to_cpu(src->members);
342
343         for (i = 0; i < 2; i++) {
344                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
345                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
346                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
347                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
348         }
349
350         dst->usr_time           = le64_to_cpu(src->usr_time);
351         dst->sys_time           = le64_to_cpu(src->sys_time);
352         dst->ctx                = le64_to_cpu(src->ctx);
353         dst->minf               = le64_to_cpu(src->minf);
354         dst->majf               = le64_to_cpu(src->majf);
355         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
356         dst->percentile_list    = NULL;
357
358         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
359                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
360                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
361                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
362         }
363
364         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
365                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
366                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
367         }
368
369         for (i = 0; i < 2; i++)
370                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
371                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
372
373         for (i = 0; i < 3; i++) {
374                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
375                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
376         }
377
378         dst->total_submit       = le64_to_cpu(src->total_submit);
379         dst->total_complete     = le64_to_cpu(src->total_complete);
380
381         for (i = 0; i < 2; i++) {
382                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
383                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
384         }
385
386         dst->total_run_time     = le64_to_cpu(src->total_run_time);
387         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
388         dst->total_err_count    = le64_to_cpu(src->total_err_count);
389         dst->first_error        = le32_to_cpu(src->first_error);
390         dst->kb_base            = le32_to_cpu(src->kb_base);
391 }
392
393 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
394 {
395         int i;
396
397         for (i = 0; i < 2; i++) {
398                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
399                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
400                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
401                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
402                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
403                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
404         }
405
406         dst->kb_base    = le32_to_cpu(src->kb_base);
407         dst->groupid    = le32_to_cpu(src->groupid);
408 }
409
410 static void handle_ts(struct fio_net_cmd *cmd)
411 {
412         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
413
414         convert_ts(&p->ts, &p->ts);
415         convert_gs(&p->rs, &p->rs);
416
417         show_thread_status(&p->ts, &p->rs);
418 }
419
420 static void handle_gs(struct fio_net_cmd *cmd)
421 {
422         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
423
424         convert_gs(gs, gs);
425         show_group_stats(gs);
426 }
427
428 static void handle_eta(struct fio_net_cmd *cmd)
429 {
430         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
431         int i;
432
433         je->nr_running          = le32_to_cpu(je->nr_running);
434         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
435         je->nr_pending          = le32_to_cpu(je->nr_pending);
436         je->files_open          = le32_to_cpu(je->files_open);
437         je->m_rate              = le32_to_cpu(je->m_rate);
438         je->t_rate              = le32_to_cpu(je->t_rate);
439         je->m_iops              = le32_to_cpu(je->m_iops);
440         je->t_iops              = le32_to_cpu(je->t_iops);
441
442         for (i = 0; i < 2; i++) {
443                 je->rate[i]     = le32_to_cpu(je->rate[i]);
444                 je->iops[i]     = le32_to_cpu(je->iops[i]);
445         }
446
447         je->elapsed_sec         = le32_to_cpu(je->nr_running);
448         je->eta_sec             = le64_to_cpu(je->eta_sec);
449
450         display_thread_status(je);
451 }
452
453 static void handle_probe(struct fio_net_cmd *cmd)
454 {
455         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
456
457         log_info("Probe: hostname=%s, fio ver %u.%u.%u\n", probe->hostname,
458                         probe->fio_major, probe->fio_minor, probe->fio_patch);
459 }
460
461 static int handle_client(struct fio_client *client, int one)
462 {
463         struct fio_net_cmd *cmd;
464         int done = 0;
465
466         dprint(FD_NET, "client: handle %s\n", client->hostname);
467
468         while ((cmd = fio_net_recv_cmd(client->fd, 1)) != NULL) {
469                 dprint(FD_NET, "%s: got cmd op %d\n", client->hostname,
470                                                         cmd->opcode);
471
472                 switch (cmd->opcode) {
473                 case FIO_NET_CMD_QUIT:
474                         remove_client(client);
475                         free(cmd);
476                         done = 1;
477                         break;
478                 case FIO_NET_CMD_TEXT: {
479                         const char *buf = (const char *) cmd->payload;
480
481                         if (!client->skip_newline)
482                                 fprintf(f_out, "Client <%s>: ", client->hostname);
483                         fwrite(buf, cmd->pdu_len, 1, f_out);
484                         fflush(f_out);
485                         client->skip_newline = strchr(buf, '\n') == NULL;
486                         free(cmd);
487                         break;
488                         }
489                 case FIO_NET_CMD_TS:
490                         handle_ts(cmd);
491                         free(cmd);
492                         break;
493                 case FIO_NET_CMD_GS:
494                         handle_gs(cmd);
495                         free(cmd);
496                         break;
497                 case FIO_NET_CMD_ETA:
498                         handle_eta(cmd);
499                         free(cmd);
500                         break;
501                 case FIO_NET_CMD_PROBE:
502                         handle_probe(cmd);
503                         free(cmd);
504                         break;
505                 case FIO_NET_CMD_START:
506                         client->state = Client_started;
507                         free(cmd);
508                         break;
509                 case FIO_NET_CMD_STOP:
510                         client->state = Client_stopped;
511                         free(cmd);
512                         break;
513                 default:
514                         log_err("fio: unknown client op: %d\n", cmd->opcode);
515                         free(cmd);
516                         break;
517                 }
518
519                 if (done || one)
520                         break;
521         }
522
523         return 0;
524 }
525
526 int fio_handle_clients(void)
527 {
528         struct fio_client *client;
529         struct flist_head *entry;
530         struct pollfd *pfds;
531         int i, ret = 0;
532
533         pfds = malloc(nr_clients * sizeof(struct pollfd));
534
535         while (!exit_backend && nr_clients) {
536                 i = 0;
537                 flist_for_each(entry, &client_list) {
538                         client = flist_entry(entry, struct fio_client, list);
539
540                         pfds[i].fd = client->fd;
541                         pfds[i].events = POLLIN;
542                         i++;
543                 }
544
545                 assert(i == nr_clients);
546
547                 do {
548                         ret = poll(pfds, nr_clients, 100);
549                         if (ret < 0) {
550                                 if (errno == EINTR)
551                                         continue;
552                                 log_err("fio: poll clients: %s\n", strerror(errno));
553                                 break;
554                         } else if (!ret)
555                                 continue;
556                 } while (ret <= 0);
557
558                 for (i = 0; i < nr_clients; i++) {
559                         if (!(pfds[i].revents & POLLIN))
560                                 continue;
561
562                         client = find_client_by_fd(pfds[i].fd);
563                         if (!client) {
564                                 log_err("fio: unknown client\n");
565                                 continue;
566                         }
567                         handle_client(client, 0);
568                 }
569         }
570
571         free(pfds);
572         return 0;
573 }