projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
client: parse env variables before sending job-file contents to server
[fio.git] / client.c
... / ...
CommitLineData
1#include <stdio.h>
2#include <stdlib.h>
3#include <unistd.h>
4#include <errno.h>
5#include <fcntl.h>
6#include <poll.h>
7#include <sys/types.h>
8#include <sys/stat.h>
9#include <sys/socket.h>
10#include <sys/un.h>
11#include <netinet/in.h>
12#include <arpa/inet.h>
13#include <netdb.h>
14#include <signal.h>
15#ifdef CONFIG_ZLIB
16#include <zlib.h>
17#endif
18
19#include "fio.h"
20#include "client.h"
21#include "server.h"
22#include "flist.h"
23#include "hash.h"
24#include "verify-state.h"
25
26static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
27static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd);
28static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd);
29static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
30static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd);
31static void handle_stop(struct fio_client *client);
32static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd);
33
34static void convert_text(struct fio_net_cmd *cmd);
35
36struct client_ops fio_client_ops = {
37 .text = handle_text,
38 .disk_util = handle_du,
39 .thread_status = handle_ts,
40 .group_stats = handle_gs,
41 .stop = handle_stop,
42 .start = handle_start,
43 .eta = display_thread_status,
44 .probe = handle_probe,
45 .eta_msec = FIO_CLIENT_DEF_ETA_MSEC,
46 .client_type = FIO_CLIENT_TYPE_CLI,
47};
48
49static struct timespec eta_ts;
50
51static FLIST_HEAD(client_list);
52static FLIST_HEAD(eta_list);
53
54static FLIST_HEAD(arg_list);
55
56struct thread_stat client_ts;
57struct group_run_stats client_gs;
58int sum_stat_clients;
59
60static int sum_stat_nr;
61static struct json_object *root = NULL;
62static struct json_object *job_opt_object = NULL;
63static struct json_array *clients_array = NULL;
64static struct json_array *du_array = NULL;
65
66static int error_clients;
67
68#define FIO_CLIENT_HASH_BITS 7
69#define FIO_CLIENT_HASH_SZ (1 << FIO_CLIENT_HASH_BITS)
70#define FIO_CLIENT_HASH_MASK (FIO_CLIENT_HASH_SZ - 1)
71static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
72
73static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *, bool *);
74
75static void fio_client_add_hash(struct fio_client *client)
76{
77 int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
78
79 bucket &= FIO_CLIENT_HASH_MASK;
80 flist_add(&client->hash_list, &client_hash[bucket]);
81}
82
83static void fio_client_remove_hash(struct fio_client *client)
84{
85 if (!flist_empty(&client->hash_list))
86 flist_del_init(&client->hash_list);
87}
88
89static void fio_init fio_client_hash_init(void)
90{
91 int i;
92
93 for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
94 INIT_FLIST_HEAD(&client_hash[i]);
95}
96
97static int read_data(int fd, void *data, size_t size)
98{
99 ssize_t ret;
100
101 while (size) {
102 ret = read(fd, data, size);
103 if (ret < 0) {
104 if (errno == EAGAIN || errno == EINTR)
105 continue;
106 break;
107 } else if (!ret)
108 break;
109 else {
110 data += ret;
111 size -= ret;
112 }
113 }
114
115 if (size)
116 return EAGAIN;
117
118 return 0;
119}
120
121static int read_ini_data(int fd, void *data, size_t size)
122{
123 char *p = data;
124 int ret = 0;
125 FILE *fp;
126
127 fp = fdopen(dup(fd), "r");
128 if (!fp)
129 return errno;
130
131 while (1) {
132 ssize_t len;
133 char buf[OPT_LEN_MAX+1], *sub;
134
135 if (!fgets(buf, sizeof(buf), fp)) {
136 if (ferror(fp)) {
137 if (errno == EAGAIN || errno == EINTR)
138 continue;
139 ret = errno;
140 }
141 break;
142 }
143
144 sub = fio_option_dup_subs(buf);
145 len = strlen(sub);
146 if (len + 1 > size) {
147 log_err("fio: no space left to read data\n");
148 free(sub);
149 ret = ENOSPC;
150 break;
151 }
152
153 memcpy(p, sub, len);
154 free(sub);
155 p += len;
156 *p = '\0';
157 size -= len;
158 }
159
160 fclose(fp);
161 return ret;
162}
163
164static void fio_client_json_init(void)
165{
166 char time_buf[32];
167 time_t time_p;
168
169 if (!(output_format & FIO_OUTPUT_JSON))
170 return;
171
172 time(&time_p);
173 os_ctime_r((const time_t *) &time_p, time_buf, sizeof(time_buf));
174 time_buf[strlen(time_buf) - 1] = '\0';
175
176 root = json_create_object();
177 json_object_add_value_string(root, "fio version", fio_version_string);
178 json_object_add_value_int(root, "timestamp", time_p);
179 json_object_add_value_string(root, "time", time_buf);
180
181 job_opt_object = json_create_object();
182 json_object_add_value_object(root, "global options", job_opt_object);
183 clients_array = json_create_array();
184 json_object_add_value_array(root, "client_stats", clients_array);
185 du_array = json_create_array();
186 json_object_add_value_array(root, "disk_util", du_array);
187}
188
189static void fio_client_json_fini(void)
190{
191 if (!(output_format & FIO_OUTPUT_JSON))
192 return;
193
194 log_info("\n");
195 json_print_object(root, NULL);
196 log_info("\n");
197 json_free_object(root);
198 root = NULL;
199 clients_array = NULL;
200 du_array = NULL;
201}
202
203static struct fio_client *find_client_by_fd(int fd)
204{
205 int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
206 struct fio_client *client;
207 struct flist_head *entry;
208
209 flist_for_each(entry, &client_hash[bucket]) {
210 client = flist_entry(entry, struct fio_client, hash_list);
211
212 if (client->fd == fd) {
213 client->refs++;
214 return client;
215 }
216 }
217
218 return NULL;
219}
220
221void fio_put_client(struct fio_client *client)
222{
223 if (--client->refs)
224 return;
225
226 free(client->hostname);
227 if (client->argv)
228 free(client->argv);
229 if (client->name)
230 free(client->name);
231 while (client->nr_files) {
232 struct client_file *cf = &client->files[--client->nr_files];
233
234 free(cf->file);
235 }
236 if (client->files)
237 free(client->files);
238 if (client->opt_lists)
239 free(client->opt_lists);
240
241 if (!client->did_stat)
242 sum_stat_clients--;
243
244 if (client->error)
245 error_clients++;
246
247 free(client);
248}
249
250static int fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
251{
252 if (!--eta->pending) {
253 eta_fn(&eta->eta);
254 free(eta);
255 return 0;
256 }
257
258 return 1;
259}
260
261static void fio_drain_client_text(struct fio_client *client)
262{
263 do {
264 struct fio_net_cmd *cmd;
265
266 cmd = fio_net_recv_cmd(client->fd, false);
267 if (!cmd)
268 break;
269
270 if (cmd->opcode == FIO_NET_CMD_TEXT) {
271 convert_text(cmd);
272 client->ops->text(client, cmd);
273 }
274
275 free(cmd);
276 } while (1);
277}
278
279static void remove_client(struct fio_client *client)
280{
281 assert(client->refs);
282
283 dprint(FD_NET, "client: removed <%s>\n", client->hostname);
284
285 fio_drain_client_text(client);
286
287 if (!flist_empty(&client->list))
288 flist_del_init(&client->list);
289
290 fio_client_remove_hash(client);
291
292 if (!flist_empty(&client->eta_list)) {
293 flist_del_init(&client->eta_list);
294 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
295 }
296
297 close(client->fd);
298 client->fd = -1;
299
300 if (client->ops->removed)
301 client->ops->removed(client);
302
303 nr_clients--;
304 fio_put_client(client);
305}
306
307struct fio_client *fio_get_client(struct fio_client *client)
308{
309 client->refs++;
310 return client;
311}
312
313static void __fio_client_add_cmd_option(struct fio_client *client,
314 const char *opt)
315{
316 int index;
317
318 index = client->argc++;
319 client->argv = realloc(client->argv, sizeof(char *) * client->argc);
320 client->argv[index] = strdup(opt);
321 dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
322}
323
324void fio_client_add_cmd_option(void *cookie, const char *opt)
325{
326 struct fio_client *client = cookie;
327 struct flist_head *entry;
328
329 if (!client || !opt)
330 return;
331
332 __fio_client_add_cmd_option(client, opt);
333
334 /*
335 * Duplicate arguments to shared client group
336 */
337 flist_for_each(entry, &arg_list) {
338 client = flist_entry(entry, struct fio_client, arg_list);
339
340 __fio_client_add_cmd_option(client, opt);
341 }
342}
343
344struct fio_client *fio_client_add_explicit(struct client_ops *ops,
345 const char *hostname, int type,
346 int port)
347{
348 struct fio_client *client;
349
350 client = malloc(sizeof(*client));
351 memset(client, 0, sizeof(*client));
352
353 INIT_FLIST_HEAD(&client->list);
354 INIT_FLIST_HEAD(&client->hash_list);
355 INIT_FLIST_HEAD(&client->arg_list);
356 INIT_FLIST_HEAD(&client->eta_list);
357 INIT_FLIST_HEAD(&client->cmd_list);
358
359 client->hostname = strdup(hostname);
360
361 if (type == Fio_client_socket)
362 client->is_sock = true;
363 else {
364 int ipv6;
365
366 ipv6 = type == Fio_client_ipv6;
367 if (fio_server_parse_host(hostname, ipv6,
368 &client->addr.sin_addr,
369 &client->addr6.sin6_addr))
370 goto err;
371
372 client->port = port;
373 }
374
375 client->fd = -1;
376 client->ops = ops;
377 client->refs = 1;
378 client->type = ops->client_type;
379
380 __fio_client_add_cmd_option(client, "fio");
381
382 flist_add(&client->list, &client_list);
383 nr_clients++;
384 dprint(FD_NET, "client: added <%s>\n", client->hostname);
385 return client;
386err:
387 free(client);
388 return NULL;
389}
390
391int fio_client_add_ini_file(void *cookie, const char *ini_file, bool remote)
392{
393 struct fio_client *client = cookie;
394 struct client_file *cf;
395 size_t new_size;
396 void *new_files;
397
398 if (!client)
399 return 1;
400
401 dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file);
402
403 new_size = (client->nr_files + 1) * sizeof(struct client_file);
404 new_files = realloc(client->files, new_size);
405 if (!new_files)
406 return 1;
407
408 client->files = new_files;
409 cf = &client->files[client->nr_files];
410 cf->file = strdup(ini_file);
411 cf->remote = remote;
412 client->nr_files++;
413 return 0;
414}
415
416int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
417{
418 struct fio_client *existing = *cookie;
419 struct fio_client *client;
420
421 if (existing) {
422 /*
423 * We always add our "exec" name as the option, hence 1
424 * means empty.
425 */
426 if (existing->argc == 1)
427 flist_add_tail(&existing->arg_list, &arg_list);
428 else {
429 while (!flist_empty(&arg_list))
430 flist_del_init(arg_list.next);
431 }
432 }
433
434 client = malloc(sizeof(*client));
435 memset(client, 0, sizeof(*client));
436
437 INIT_FLIST_HEAD(&client->list);
438 INIT_FLIST_HEAD(&client->hash_list);
439 INIT_FLIST_HEAD(&client->arg_list);
440 INIT_FLIST_HEAD(&client->eta_list);
441 INIT_FLIST_HEAD(&client->cmd_list);
442
443 if (fio_server_parse_string(hostname, &client->hostname,
444 &client->is_sock, &client->port,
445 &client->addr.sin_addr,
446 &client->addr6.sin6_addr,
447 &client->ipv6))
448 return -1;
449
450 client->fd = -1;
451 client->ops = ops;
452 client->refs = 1;
453 client->type = ops->client_type;
454
455 __fio_client_add_cmd_option(client, "fio");
456
457 flist_add(&client->list, &client_list);
458 nr_clients++;
459 dprint(FD_NET, "client: added <%s>\n", client->hostname);
460 *cookie = client;
461 return 0;
462}
463
464static const char *server_name(struct fio_client *client, char *buf,
465 size_t bufsize)
466{
467 const char *from;
468
469 if (client->ipv6)
470 from = inet_ntop(AF_INET6, (struct sockaddr *) &client->addr6.sin6_addr, buf, bufsize);
471 else if (client->is_sock)
472 from = "sock";
473 else
474 from = inet_ntop(AF_INET, (struct sockaddr *) &client->addr.sin_addr, buf, bufsize);
475
476 return from;
477}
478
479static void probe_client(struct fio_client *client)
480{
481 struct cmd_client_probe_pdu pdu;
482 const char *sname;
483 uint64_t tag;
484 char buf[64];
485
486 dprint(FD_NET, "client: send probe\n");
487
488#ifdef CONFIG_ZLIB
489 pdu.flags = __le64_to_cpu(FIO_PROBE_FLAG_ZLIB);
490#else
491 pdu.flags = 0;
492#endif
493
494 sname = server_name(client, buf, sizeof(buf));
495 memset(pdu.server, 0, sizeof(pdu.server));
496 strncpy((char *) pdu.server, sname, sizeof(pdu.server) - 1);
497
498 fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list);
499}
500
501static int fio_client_connect_ip(struct fio_client *client)
502{
503 struct sockaddr *addr;
504 socklen_t socklen;
505 int fd, domain;
506
507 if (client->ipv6) {
508 client->addr6.sin6_family = AF_INET6;
509 client->addr6.sin6_port = htons(client->port);
510 domain = AF_INET6;
511 addr = (struct sockaddr *) &client->addr6;
512 socklen = sizeof(client->addr6);
513 } else {
514 client->addr.sin_family = AF_INET;
515 client->addr.sin_port = htons(client->port);
516 domain = AF_INET;
517 addr = (struct sockaddr *) &client->addr;
518 socklen = sizeof(client->addr);
519 }
520
521 fd = socket(domain, SOCK_STREAM, 0);
522 if (fd < 0) {
523 int ret = -errno;
524
525 log_err("fio: socket: %s\n", strerror(errno));
526 return ret;
527 }
528
529 if (connect(fd, addr, socklen) < 0) {
530 int ret = -errno;
531
532 log_err("fio: connect: %s\n", strerror(errno));
533 log_err("fio: failed to connect to %s:%u\n", client->hostname,
534 client->port);
535 close(fd);
536 return ret;
537 }
538
539 return fd;
540}
541
542static int fio_client_connect_sock(struct fio_client *client)
543{
544 struct sockaddr_un *addr = &client->addr_un;
545 socklen_t len;
546 int fd;
547
548 memset(addr, 0, sizeof(*addr));
549 addr->sun_family = AF_UNIX;
550 strncpy(addr->sun_path, client->hostname, sizeof(addr->sun_path) - 1);
551
552 fd = socket(AF_UNIX, SOCK_STREAM, 0);
553 if (fd < 0) {
554 int ret = -errno;
555
556 log_err("fio: socket: %s\n", strerror(errno));
557 return ret;
558 }
559
560 len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
561 if (connect(fd, (struct sockaddr *) addr, len) < 0) {
562 int ret = -errno;
563
564 log_err("fio: connect; %s\n", strerror(errno));
565 close(fd);
566 return ret;
567 }
568
569 return fd;
570}
571
572int fio_client_connect(struct fio_client *client)
573{
574 int fd;
575
576 dprint(FD_NET, "client: connect to host %s\n", client->hostname);
577
578 if (client->is_sock)
579 fd = fio_client_connect_sock(client);
580 else
581 fd = fio_client_connect_ip(client);
582
583 dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
584
585 if (fd < 0)
586 return fd;
587
588 client->fd = fd;
589 fio_client_add_hash(client);
590 client->state = Client_connected;
591
592 probe_client(client);
593 return 0;
594}
595
596int fio_client_terminate(struct fio_client *client)
597{
598 return fio_net_send_quit(client->fd);
599}
600
601static void fio_clients_terminate(void)
602{
603 struct flist_head *entry;
604 struct fio_client *client;
605
606 dprint(FD_NET, "client: terminate clients\n");
607
608 flist_for_each(entry, &client_list) {
609 client = flist_entry(entry, struct fio_client, list);
610 fio_client_terminate(client);
611 }
612}
613
614static void sig_int(int sig)
615{
616 dprint(FD_NET, "client: got signal %d\n", sig);
617 fio_clients_terminate();
618}
619
620static void client_signal_handler(void)
621{
622 struct sigaction act;
623
624 memset(&act, 0, sizeof(act));
625 act.sa_handler = sig_int;
626 act.sa_flags = SA_RESTART;
627 sigaction(SIGINT, &act, NULL);
628
629 memset(&act, 0, sizeof(act));
630 act.sa_handler = sig_int;
631 act.sa_flags = SA_RESTART;
632 sigaction(SIGTERM, &act, NULL);
633
634/* Windows uses SIGBREAK as a quit signal from other applications */
635#ifdef WIN32
636 memset(&act, 0, sizeof(act));
637 act.sa_handler = sig_int;
638 act.sa_flags = SA_RESTART;
639 sigaction(SIGBREAK, &act, NULL);
640#endif
641
642 memset(&act, 0, sizeof(act));
643 act.sa_handler = sig_show_status;
644 act.sa_flags = SA_RESTART;
645 sigaction(SIGUSR1, &act, NULL);
646}
647
648static int send_client_cmd_line(struct fio_client *client)
649{
650 struct cmd_single_line_pdu *cslp;
651 struct cmd_line_pdu *clp;
652 unsigned long offset;
653 unsigned int *lens;
654 void *pdu;
655 size_t mem;
656 int i, ret;
657
658 dprint(FD_NET, "client: send cmdline %d\n", client->argc);
659
660 lens = malloc(client->argc * sizeof(unsigned int));
661
662 /*
663 * Find out how much mem we need
664 */
665 for (i = 0, mem = 0; i < client->argc; i++) {
666 lens[i] = strlen(client->argv[i]) + 1;
667 mem += lens[i];
668 }
669
670 /*
671 * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
672 */
673 mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
674
675 pdu = malloc(mem);
676 clp = pdu;
677 offset = sizeof(*clp);
678
679 for (i = 0; i < client->argc; i++) {
680 uint16_t arg_len = lens[i];
681
682 cslp = pdu + offset;
683 strcpy((char *) cslp->text, client->argv[i]);
684 cslp->len = cpu_to_le16(arg_len);
685 offset += sizeof(*cslp) + arg_len;
686 }
687
688 free(lens);
689 clp->lines = cpu_to_le16(client->argc);
690 clp->client_type = __cpu_to_le16(client->type);
691 ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, NULL, NULL);
692 free(pdu);
693 return ret;
694}
695
696int fio_clients_connect(void)
697{
698 struct fio_client *client;
699 struct flist_head *entry, *tmp;
700 int ret;
701
702#ifdef WIN32
703 WSADATA wsd;
704 WSAStartup(MAKEWORD(2, 2), &wsd);
705#endif
706
707 dprint(FD_NET, "client: connect all\n");
708
709 client_signal_handler();
710
711 flist_for_each_safe(entry, tmp, &client_list) {
712 client = flist_entry(entry, struct fio_client, list);
713
714 ret = fio_client_connect(client);
715 if (ret) {
716 remove_client(client);
717 continue;
718 }
719
720 if (client->argc > 1)
721 send_client_cmd_line(client);
722 }
723
724 return !nr_clients;
725}
726
727int fio_start_client(struct fio_client *client)
728{
729 dprint(FD_NET, "client: start %s\n", client->hostname);
730 return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
731}
732
733int fio_start_all_clients(void)
734{
735 struct fio_client *client;
736 struct flist_head *entry, *tmp;
737 int ret;
738
739 dprint(FD_NET, "client: start all\n");
740
741 fio_client_json_init();
742
743 flist_for_each_safe(entry, tmp, &client_list) {
744 client = flist_entry(entry, struct fio_client, list);
745
746 ret = fio_start_client(client);
747 if (ret) {
748 remove_client(client);
749 continue;
750 }
751 }
752
753 return flist_empty(&client_list);
754}
755
756static int __fio_client_send_remote_ini(struct fio_client *client,
757 const char *filename)
758{
759 struct cmd_load_file_pdu *pdu;
760 size_t p_size;
761 int ret;
762
763 dprint(FD_NET, "send remote ini %s to %s\n", filename, client->hostname);
764
765 p_size = sizeof(*pdu) + strlen(filename) + 1;
766 pdu = malloc(p_size);
767 memset(pdu, 0, p_size);
768 pdu->name_len = strlen(filename);
769 strcpy((char *) pdu->file, filename);
770 pdu->client_type = cpu_to_le16((uint16_t) client->type);
771
772 client->sent_job = true;
773 ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_LOAD_FILE, pdu, p_size,NULL, NULL);
774 free(pdu);
775 return ret;
776}
777
778/*
779 * Send file contents to server backend. We could use sendfile(), but to remain
780 * more portable lets just read/write the darn thing.
781 */
782static int __fio_client_send_local_ini(struct fio_client *client,
783 const char *filename)
784{
785 struct cmd_job_pdu *pdu;
786 size_t p_size;
787 struct stat sb;
788 char *p;
789 void *buf;
790 off_t len;
791 int fd, ret;
792
793 dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
794
795 fd = open(filename, O_RDONLY);
796 if (fd < 0) {
797 ret = -errno;
798 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
799 return ret;
800 }
801
802 if (fstat(fd, &sb) < 0) {
803 ret = -errno;
804 log_err("fio: job file stat: %s\n", strerror(errno));
805 close(fd);
806 return ret;
807 }
808
809 /*
810 * Add extra space for variable expansion, but doesn't guarantee.
811 */
812 sb.st_size += OPT_LEN_MAX;
813 p_size = sb.st_size + sizeof(*pdu);
814 pdu = malloc(p_size);
815 buf = pdu->buf;
816
817 len = sb.st_size;
818 p = buf;
819 if (read_ini_data(fd, p, len)) {
820 log_err("fio: failed reading job file %s\n", filename);
821 close(fd);
822 free(pdu);
823 return 1;
824 }
825
826 pdu->buf_len = __cpu_to_le32(sb.st_size);
827 pdu->client_type = cpu_to_le32(client->type);
828
829 client->sent_job = true;
830 ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, NULL, NULL);
831 free(pdu);
832 close(fd);
833 return ret;
834}
835
836int fio_client_send_ini(struct fio_client *client, const char *filename,
837 bool remote)
838{
839 int ret;
840
841 if (!remote)
842 ret = __fio_client_send_local_ini(client, filename);
843 else
844 ret = __fio_client_send_remote_ini(client, filename);
845
846 if (!ret)
847 client->sent_job = true;
848
849 return ret;
850}
851
852static int fio_client_send_cf(struct fio_client *client,
853 struct client_file *cf)
854{
855 return fio_client_send_ini(client, cf->file, cf->remote);
856}
857
858int fio_clients_send_ini(const char *filename)
859{
860 struct fio_client *client;
861 struct flist_head *entry, *tmp;
862
863 flist_for_each_safe(entry, tmp, &client_list) {
864 bool failed = false;
865
866 client = flist_entry(entry, struct fio_client, list);
867
868 if (client->nr_files) {
869 int i;
870
871 for (i = 0; i < client->nr_files; i++) {
872 struct client_file *cf;
873
874 cf = &client->files[i];
875
876 if (fio_client_send_cf(client, cf)) {
877 failed = true;
878 remove_client(client);
879 break;
880 }
881 }
882 }
883 if (client->sent_job || failed)
884 continue;
885 if (!filename || fio_client_send_ini(client, filename, 0))
886 remove_client(client);
887 }
888
889 return !nr_clients;
890}
891
892int fio_client_update_options(struct fio_client *client,
893 struct thread_options *o, uint64_t *tag)
894{
895 struct cmd_add_job_pdu pdu;
896
897 pdu.thread_number = cpu_to_le32(client->thread_number);
898 pdu.groupid = cpu_to_le32(client->groupid);
899 convert_thread_options_to_net(&pdu.top, o);
900
901 return fio_net_send_cmd(client->fd, FIO_NET_CMD_UPDATE_JOB, &pdu, sizeof(pdu), tag, &client->cmd_list);
902}
903
904static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
905{
906 dst->max_val = le64_to_cpu(src->max_val);
907 dst->min_val = le64_to_cpu(src->min_val);
908 dst->samples = le64_to_cpu(src->samples);
909
910 /*
911 * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
912 */
913 dst->mean.u.f = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
914 dst->S.u.f = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
915}
916
917static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
918{
919 int i, j;
920
921 dst->error = le32_to_cpu(src->error);
922 dst->thread_number = le32_to_cpu(src->thread_number);
923 dst->groupid = le32_to_cpu(src->groupid);
924 dst->pid = le32_to_cpu(src->pid);
925 dst->members = le32_to_cpu(src->members);
926 dst->unified_rw_rep = le32_to_cpu(src->unified_rw_rep);
927
928 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
929 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
930 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
931 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
932 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
933 convert_io_stat(&dst->iops_stat[i], &src->iops_stat[i]);
934 }
935
936 dst->usr_time = le64_to_cpu(src->usr_time);
937 dst->sys_time = le64_to_cpu(src->sys_time);
938 dst->ctx = le64_to_cpu(src->ctx);
939 dst->minf = le64_to_cpu(src->minf);
940 dst->majf = le64_to_cpu(src->majf);
941 dst->clat_percentiles = le32_to_cpu(src->clat_percentiles);
942 dst->lat_percentiles = le32_to_cpu(src->lat_percentiles);
943 dst->percentile_precision = le64_to_cpu(src->percentile_precision);
944
945 for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
946 fio_fp64_t *fps = &src->percentile_list[i];
947 fio_fp64_t *fpd = &dst->percentile_list[i];
948
949 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
950 }
951
952 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
953 dst->io_u_map[i] = le64_to_cpu(src->io_u_map[i]);
954 dst->io_u_submit[i] = le64_to_cpu(src->io_u_submit[i]);
955 dst->io_u_complete[i] = le64_to_cpu(src->io_u_complete[i]);
956 }
957
958 for (i = 0; i < FIO_IO_U_LAT_N_NR; i++)
959 dst->io_u_lat_n[i] = le64_to_cpu(src->io_u_lat_n[i]);
960 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
961 dst->io_u_lat_u[i] = le64_to_cpu(src->io_u_lat_u[i]);
962 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
963 dst->io_u_lat_m[i] = le64_to_cpu(src->io_u_lat_m[i]);
964
965 for (i = 0; i < DDIR_RWDIR_CNT; i++)
966 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
967 dst->io_u_plat[i][j] = le64_to_cpu(src->io_u_plat[i][j]);
968
969 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
970 dst->total_io_u[i] = le64_to_cpu(src->total_io_u[i]);
971 dst->short_io_u[i] = le64_to_cpu(src->short_io_u[i]);
972 dst->drop_io_u[i] = le64_to_cpu(src->drop_io_u[i]);
973 }
974
975 dst->total_submit = le64_to_cpu(src->total_submit);
976 dst->total_complete = le64_to_cpu(src->total_complete);
977
978 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
979 dst->io_bytes[i] = le64_to_cpu(src->io_bytes[i]);
980 dst->runtime[i] = le64_to_cpu(src->runtime[i]);
981 }
982
983 dst->total_run_time = le64_to_cpu(src->total_run_time);
984 dst->continue_on_error = le16_to_cpu(src->continue_on_error);
985 dst->total_err_count = le64_to_cpu(src->total_err_count);
986 dst->first_error = le32_to_cpu(src->first_error);
987 dst->kb_base = le32_to_cpu(src->kb_base);
988 dst->unit_base = le32_to_cpu(src->unit_base);
989
990 dst->sig_figs = le32_to_cpu(src->sig_figs);
991
992 dst->latency_depth = le32_to_cpu(src->latency_depth);
993 dst->latency_target = le64_to_cpu(src->latency_target);
994 dst->latency_window = le64_to_cpu(src->latency_window);
995 dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
996
997 dst->nr_block_infos = le64_to_cpu(src->nr_block_infos);
998 for (i = 0; i < dst->nr_block_infos; i++)
999 dst->block_infos[i] = le32_to_cpu(src->block_infos[i]);
1000
1001 dst->ss_dur = le64_to_cpu(src->ss_dur);
1002 dst->ss_state = le32_to_cpu(src->ss_state);
1003 dst->ss_head = le32_to_cpu(src->ss_head);
1004 dst->ss_limit.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_limit.u.i));
1005 dst->ss_slope.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_slope.u.i));
1006 dst->ss_deviation.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_deviation.u.i));
1007 dst->ss_criterion.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_criterion.u.i));
1008
1009 if (dst->ss_state & FIO_SS_DATA) {
1010 for (i = 0; i < dst->ss_dur; i++ ) {
1011 dst->ss_iops_data[i] = le64_to_cpu(src->ss_iops_data[i]);
1012 dst->ss_bw_data[i] = le64_to_cpu(src->ss_bw_data[i]);
1013 }
1014 }
1015}
1016
1017static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
1018{
1019 int i;
1020
1021 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1022 dst->max_run[i] = le64_to_cpu(src->max_run[i]);
1023 dst->min_run[i] = le64_to_cpu(src->min_run[i]);
1024 dst->max_bw[i] = le64_to_cpu(src->max_bw[i]);
1025 dst->min_bw[i] = le64_to_cpu(src->min_bw[i]);
1026 dst->iobytes[i] = le64_to_cpu(src->iobytes[i]);
1027 dst->agg[i] = le64_to_cpu(src->agg[i]);
1028 }
1029
1030 dst->kb_base = le32_to_cpu(src->kb_base);
1031 dst->unit_base = le32_to_cpu(src->unit_base);
1032 dst->sig_figs = le32_to_cpu(src->sig_figs);
1033 dst->groupid = le32_to_cpu(src->groupid);
1034 dst->unified_rw_rep = le32_to_cpu(src->unified_rw_rep);
1035}
1036
1037static void json_object_add_client_info(struct json_object *obj,
1038 struct fio_client *client)
1039{
1040 const char *hostname = client->hostname ? client->hostname : "";
1041
1042 json_object_add_value_string(obj, "hostname", hostname);
1043 json_object_add_value_int(obj, "port", client->port);
1044}
1045
1046static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
1047{
1048 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1049 struct flist_head *opt_list = NULL;
1050 struct json_object *tsobj;
1051
1052 if (client->opt_lists && p->ts.thread_number <= client->jobs)
1053 opt_list = &client->opt_lists[p->ts.thread_number - 1];
1054
1055 tsobj = show_thread_status(&p->ts, &p->rs, opt_list, NULL);
1056 client->did_stat = true;
1057 if (tsobj) {
1058 json_object_add_client_info(tsobj, client);
1059 json_array_add_value_object(clients_array, tsobj);
1060 }
1061
1062 if (sum_stat_clients <= 1)
1063 return;
1064
1065 sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1);
1066 sum_group_stats(&client_gs, &p->rs);
1067
1068 client_ts.members++;
1069 client_ts.thread_number = p->ts.thread_number;
1070 client_ts.groupid = p->ts.groupid;
1071 client_ts.unified_rw_rep = p->ts.unified_rw_rep;
1072 client_ts.sig_figs = p->ts.sig_figs;
1073
1074 if (++sum_stat_nr == sum_stat_clients) {
1075 strcpy(client_ts.name, "All clients");
1076 tsobj = show_thread_status(&client_ts, &client_gs, NULL, NULL);
1077 if (tsobj) {
1078 json_object_add_client_info(tsobj, client);
1079 json_array_add_value_object(clients_array, tsobj);
1080 }
1081 }
1082}
1083
1084static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
1085{
1086 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1087
1088 if (output_format & FIO_OUTPUT_NORMAL)
1089 show_group_stats(gs, NULL);
1090}
1091
1092static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
1093{
1094 struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
1095 struct print_option *p;
1096
1097 if (!job_opt_object)
1098 return;
1099
1100 pdu->global = le16_to_cpu(pdu->global);
1101 pdu->truncated = le16_to_cpu(pdu->truncated);
1102 pdu->groupid = le32_to_cpu(pdu->groupid);
1103
1104 p = malloc(sizeof(*p));
1105 p->name = strdup((char *) pdu->name);
1106 if (pdu->value[0] != '\0')
1107 p->value = strdup((char *) pdu->value);
1108 else
1109 p->value = NULL;
1110
1111 if (pdu->global) {
1112 const char *pos = "";
1113
1114 if (p->value)
1115 pos = p->value;
1116
1117 json_object_add_value_string(job_opt_object, p->name, pos);
1118 } else if (client->opt_lists) {
1119 struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
1120
1121 flist_add_tail(&p->list, opt_list);
1122 }
1123}
1124
1125static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
1126{
1127 struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1128 const char *buf = (const char *) pdu->buf;
1129 const char *name;
1130 int fio_unused ret;
1131
1132 name = client->name ? client->name : client->hostname;
1133
1134 if (!client->skip_newline)
1135 fprintf(f_out, "<%s> ", name);
1136 ret = fwrite(buf, pdu->buf_len, 1, f_out);
1137 fflush(f_out);
1138 client->skip_newline = strchr(buf, '\n') == NULL;
1139}
1140
1141static void convert_agg(struct disk_util_agg *agg)
1142{
1143 int i;
1144
1145 for (i = 0; i < 2; i++) {
1146 agg->ios[i] = le64_to_cpu(agg->ios[i]);
1147 agg->merges[i] = le64_to_cpu(agg->merges[i]);
1148 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
1149 agg->ticks[i] = le64_to_cpu(agg->ticks[i]);
1150 }
1151
1152 agg->io_ticks = le64_to_cpu(agg->io_ticks);
1153 agg->time_in_queue = le64_to_cpu(agg->time_in_queue);
1154 agg->slavecount = le32_to_cpu(agg->slavecount);
1155 agg->max_util.u.f = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
1156}
1157
1158static void convert_dus(struct disk_util_stat *dus)
1159{
1160 int i;
1161
1162 for (i = 0; i < 2; i++) {
1163 dus->s.ios[i] = le64_to_cpu(dus->s.ios[i]);
1164 dus->s.merges[i] = le64_to_cpu(dus->s.merges[i]);
1165 dus->s.sectors[i] = le64_to_cpu(dus->s.sectors[i]);
1166 dus->s.ticks[i] = le64_to_cpu(dus->s.ticks[i]);
1167 }
1168
1169 dus->s.io_ticks = le64_to_cpu(dus->s.io_ticks);
1170 dus->s.time_in_queue = le64_to_cpu(dus->s.time_in_queue);
1171 dus->s.msec = le64_to_cpu(dus->s.msec);
1172}
1173
1174static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
1175{
1176 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1177
1178 if (!client->disk_stats_shown) {
1179 client->disk_stats_shown = true;
1180 log_info("\nDisk stats (read/write):\n");
1181 }
1182
1183 if (output_format & FIO_OUTPUT_JSON) {
1184 struct json_object *duobj;
1185 json_array_add_disk_util(&du->dus, &du->agg, du_array);
1186 duobj = json_array_last_value_object(du_array);
1187 json_object_add_client_info(duobj, client);
1188 }
1189 if (output_format & FIO_OUTPUT_TERSE)
1190 print_disk_util(&du->dus, &du->agg, 1, NULL);
1191 if (output_format & FIO_OUTPUT_NORMAL)
1192 print_disk_util(&du->dus, &du->agg, 0, NULL);
1193}
1194
1195static void convert_jobs_eta(struct jobs_eta *je)
1196{
1197 int i;
1198
1199 je->nr_running = le32_to_cpu(je->nr_running);
1200 je->nr_ramp = le32_to_cpu(je->nr_ramp);
1201 je->nr_pending = le32_to_cpu(je->nr_pending);
1202 je->nr_setting_up = le32_to_cpu(je->nr_setting_up);
1203 je->files_open = le32_to_cpu(je->files_open);
1204
1205 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1206 je->m_rate[i] = le64_to_cpu(je->m_rate[i]);
1207 je->t_rate[i] = le64_to_cpu(je->t_rate[i]);
1208 je->m_iops[i] = le32_to_cpu(je->m_iops[i]);
1209 je->t_iops[i] = le32_to_cpu(je->t_iops[i]);
1210 je->rate[i] = le64_to_cpu(je->rate[i]);
1211 je->iops[i] = le32_to_cpu(je->iops[i]);
1212 }
1213
1214 je->elapsed_sec = le64_to_cpu(je->elapsed_sec);
1215 je->eta_sec = le64_to_cpu(je->eta_sec);
1216 je->nr_threads = le32_to_cpu(je->nr_threads);
1217 je->is_pow2 = le32_to_cpu(je->is_pow2);
1218 je->unit_base = le32_to_cpu(je->unit_base);
1219 je->sig_figs = le32_to_cpu(je->sig_figs);
1220}
1221
1222void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
1223{
1224 int i;
1225
1226 dst->nr_running += je->nr_running;
1227 dst->nr_ramp += je->nr_ramp;
1228 dst->nr_pending += je->nr_pending;
1229 dst->nr_setting_up += je->nr_setting_up;
1230 dst->files_open += je->files_open;
1231
1232 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1233 dst->m_rate[i] += je->m_rate[i];
1234 dst->t_rate[i] += je->t_rate[i];
1235 dst->m_iops[i] += je->m_iops[i];
1236 dst->t_iops[i] += je->t_iops[i];
1237 dst->rate[i] += je->rate[i];
1238 dst->iops[i] += je->iops[i];
1239 }
1240
1241 dst->elapsed_sec += je->elapsed_sec;
1242
1243 if (je->eta_sec > dst->eta_sec)
1244 dst->eta_sec = je->eta_sec;
1245
1246 dst->nr_threads += je->nr_threads;
1247
1248 /*
1249 * This wont be correct for multiple strings, but at least it
1250 * works for the basic cases.
1251 */
1252 strcpy((char *) dst->run_str, (char *) je->run_str);
1253}
1254
1255static bool remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
1256{
1257 struct fio_net_cmd_reply *reply = NULL;
1258 struct flist_head *entry;
1259
1260 flist_for_each(entry, &client->cmd_list) {
1261 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1262
1263 if (cmd->tag == (uintptr_t) reply)
1264 break;
1265
1266 reply = NULL;
1267 }
1268
1269 if (!reply) {
1270 log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag);
1271 return false;
1272 }
1273
1274 flist_del(&reply->list);
1275 cmd->tag = reply->saved_tag;
1276 free(reply);
1277 return true;
1278}
1279
1280int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
1281{
1282 do {
1283 struct fio_net_cmd_reply *reply = NULL;
1284 struct flist_head *entry;
1285
1286 flist_for_each(entry, &client->cmd_list) {
1287 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1288
1289 if (tag == (uintptr_t) reply)
1290 break;
1291
1292 reply = NULL;
1293 }
1294
1295 if (!reply)
1296 break;
1297
1298 usleep(1000);
1299 } while (1);
1300
1301 return 0;
1302}
1303
1304static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
1305{
1306 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1307 struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
1308
1309 dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
1310
1311 assert(client->eta_in_flight == eta);
1312
1313 client->eta_in_flight = NULL;
1314 flist_del_init(&client->eta_list);
1315 client->eta_timeouts = 0;
1316
1317 if (client->ops->jobs_eta)
1318 client->ops->jobs_eta(client, je);
1319
1320 fio_client_sum_jobs_eta(&eta->eta, je);
1321 fio_client_dec_jobs_eta(eta, client->ops->eta);
1322}
1323
1324static void client_flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
1325 uint64_t sample_size)
1326{
1327 struct io_sample *s;
1328 int log_offset;
1329 uint64_t i, j, nr_samples;
1330 struct io_u_plat_entry *entry;
1331 uint64_t *io_u_plat;
1332
1333 int stride = 1 << hist_coarseness;
1334
1335 if (!sample_size)
1336 return;
1337
1338 s = __get_sample(samples, 0, 0);
1339 log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
1340
1341 nr_samples = sample_size / __log_entry_sz(log_offset);
1342
1343 for (i = 0; i < nr_samples; i++) {
1344
1345 s = (struct io_sample *)((char *)__get_sample(samples, log_offset, i) +
1346 i * sizeof(struct io_u_plat_entry));
1347
1348 entry = s->data.plat_entry;
1349 io_u_plat = entry->io_u_plat;
1350
1351 fprintf(f, "%lu, %u, %u, ", (unsigned long) s->time,
1352 io_sample_ddir(s), s->bs);
1353 for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) {
1354 fprintf(f, "%llu, ", (unsigned long long)hist_sum(j, stride, io_u_plat, NULL));
1355 }
1356 fprintf(f, "%llu\n", (unsigned long long)
1357 hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat, NULL));
1358
1359 }
1360}
1361
1362static int fio_client_handle_iolog(struct fio_client *client,
1363 struct fio_net_cmd *cmd)
1364{
1365 struct cmd_iolog_pdu *pdu = NULL;
1366 bool store_direct;
1367 char *log_pathname = NULL;
1368 int ret = 0;
1369
1370 pdu = convert_iolog(cmd, &store_direct);
1371 if (!pdu) {
1372 log_err("fio: failed converting IO log\n");
1373 ret = 1;
1374 goto out;
1375 }
1376
1377 /* allocate buffer big enough for next sprintf() call */
1378 log_pathname = malloc(10 + strlen((char *)pdu->name) +
1379 strlen(client->hostname));
1380 if (!log_pathname) {
1381 log_err("fio: memory allocation of unique pathname failed\n");
1382 ret = -1;
1383 goto out;
1384 }
1385 /* generate a unique pathname for the log file using hostname */
1386 sprintf(log_pathname, "%s.%s", pdu->name, client->hostname);
1387
1388 if (store_direct) {
1389 ssize_t wrote;
1390 size_t sz;
1391 int fd;
1392
1393 fd = open((const char *) log_pathname,
1394 O_WRONLY | O_CREAT | O_TRUNC, 0644);
1395 if (fd < 0) {
1396 log_err("fio: open log %s: %s\n",
1397 log_pathname, strerror(errno));
1398 ret = 1;
1399 goto out;
1400 }
1401
1402 sz = cmd->pdu_len - sizeof(*pdu);
1403 wrote = write(fd, pdu->samples, sz);
1404 close(fd);
1405
1406 if (wrote != sz) {
1407 log_err("fio: short write on compressed log\n");
1408 ret = 1;
1409 goto out;
1410 }
1411
1412 ret = 0;
1413 } else {
1414 FILE *f;
1415 f = fopen((const char *) log_pathname, "w");
1416 if (!f) {
1417 log_err("fio: fopen log %s : %s\n",
1418 log_pathname, strerror(errno));
1419 ret = 1;
1420 goto out;
1421 }
1422
1423 if (pdu->log_type == IO_LOG_TYPE_HIST) {
1424 client_flush_hist_samples(f, pdu->log_hist_coarseness, pdu->samples,
1425 pdu->nr_samples * sizeof(struct io_sample));
1426 } else {
1427 flush_samples(f, pdu->samples,
1428 pdu->nr_samples * sizeof(struct io_sample));
1429 }
1430 fclose(f);
1431 ret = 0;
1432 }
1433
1434out:
1435 if (pdu && pdu != (void *) cmd->payload)
1436 free(pdu);
1437
1438 if (log_pathname)
1439 free(log_pathname);
1440
1441 return ret;
1442}
1443
1444static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
1445{
1446 struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
1447 const char *os, *arch;
1448 char bit[16];
1449
1450 os = fio_get_os_string(probe->os);
1451 if (!os)
1452 os = "unknown";
1453
1454 arch = fio_get_arch_string(probe->arch);
1455 if (!arch)
1456 os = "unknown";
1457
1458 sprintf(bit, "%d-bit", probe->bpp * 8);
1459 probe->flags = le64_to_cpu(probe->flags);
1460
1461 log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n",
1462 probe->hostname, probe->bigendian, bit, os, arch,
1463 probe->fio_version, (unsigned long) probe->flags);
1464
1465 if (!client->name)
1466 client->name = strdup((char *) probe->hostname);
1467}
1468
1469static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
1470{
1471 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1472
1473 client->state = Client_started;
1474 client->jobs = le32_to_cpu(pdu->jobs);
1475 client->nr_stat = le32_to_cpu(pdu->stat_outputs);
1476
1477 if (client->jobs) {
1478 int i;
1479
1480 if (client->opt_lists)
1481 free(client->opt_lists);
1482
1483 client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
1484 for (i = 0; i < client->jobs; i++)
1485 INIT_FLIST_HEAD(&client->opt_lists[i]);
1486 }
1487
1488 sum_stat_clients += client->nr_stat;
1489}
1490
1491static void handle_stop(struct fio_client *client)
1492{
1493 if (client->error)
1494 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
1495}
1496
1497static void convert_stop(struct fio_net_cmd *cmd)
1498{
1499 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1500
1501 pdu->error = le32_to_cpu(pdu->error);
1502}
1503
1504static void convert_text(struct fio_net_cmd *cmd)
1505{
1506 struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1507
1508 pdu->level = le32_to_cpu(pdu->level);
1509 pdu->buf_len = le32_to_cpu(pdu->buf_len);
1510 pdu->log_sec = le64_to_cpu(pdu->log_sec);
1511 pdu->log_usec = le64_to_cpu(pdu->log_usec);
1512}
1513
1514static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
1515 struct cmd_iolog_pdu *pdu)
1516{
1517#ifdef CONFIG_ZLIB
1518 struct cmd_iolog_pdu *ret;
1519 z_stream stream;
1520 uint32_t nr_samples;
1521 size_t total;
1522 char *p;
1523
1524 stream.zalloc = Z_NULL;
1525 stream.zfree = Z_NULL;
1526 stream.opaque = Z_NULL;
1527 stream.avail_in = 0;
1528 stream.next_in = Z_NULL;
1529
1530 if (inflateInit(&stream) != Z_OK)
1531 return NULL;
1532
1533 /*
1534 * Get header first, it's not compressed
1535 */
1536 nr_samples = le64_to_cpu(pdu->nr_samples);
1537
1538 if (pdu->log_type == IO_LOG_TYPE_HIST)
1539 total = nr_samples * (__log_entry_sz(le32_to_cpu(pdu->log_offset)) +
1540 sizeof(struct io_u_plat_entry));
1541 else
1542 total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
1543 ret = malloc(total + sizeof(*pdu));
1544 ret->nr_samples = nr_samples;
1545
1546 memcpy(ret, pdu, sizeof(*pdu));
1547
1548 p = (char *) ret + sizeof(*pdu);
1549
1550 stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1551 stream.next_in = (void *)((char *) pdu + sizeof(*pdu));
1552 while (stream.avail_in) {
1553 unsigned int this_chunk = 65536;
1554 unsigned int this_len;
1555 int err;
1556
1557 if (this_chunk > total)
1558 this_chunk = total;
1559
1560 stream.avail_out = this_chunk;
1561 stream.next_out = (void *)p;
1562 err = inflate(&stream, Z_NO_FLUSH);
1563 /* may be Z_OK, or Z_STREAM_END */
1564 if (err < 0) {
1565 log_err("fio: inflate error %d\n", err);
1566 free(ret);
1567 ret = NULL;
1568 goto err;
1569 }
1570
1571 this_len = this_chunk - stream.avail_out;
1572 p += this_len;
1573 total -= this_len;
1574 }
1575
1576err:
1577 inflateEnd(&stream);
1578 return ret;
1579#else
1580 return NULL;
1581#endif
1582}
1583
1584/*
1585 * This has been compressed on the server side, since it can be big.
1586 * Uncompress here.
1587 */
1588static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd,
1589 bool *store_direct)
1590{
1591 struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1592 struct cmd_iolog_pdu *ret;
1593 uint64_t i;
1594 int compressed;
1595 void *samples;
1596
1597 *store_direct = false;
1598
1599 /*
1600 * Convert if compressed and we support it. If it's not
1601 * compressed, we need not do anything.
1602 */
1603 compressed = le32_to_cpu(pdu->compressed);
1604 if (compressed == XMIT_COMPRESSED) {
1605#ifndef CONFIG_ZLIB
1606 log_err("fio: server sent compressed data by mistake\n");
1607 return NULL;
1608#endif
1609 ret = convert_iolog_gz(cmd, pdu);
1610 if (!ret) {
1611 log_err("fio: failed decompressing log\n");
1612 return NULL;
1613 }
1614 } else if (compressed == STORE_COMPRESSED) {
1615 *store_direct = true;
1616 ret = pdu;
1617 } else
1618 ret = pdu;
1619
1620 ret->nr_samples = le64_to_cpu(ret->nr_samples);
1621 ret->thread_number = le32_to_cpu(ret->thread_number);
1622 ret->log_type = le32_to_cpu(ret->log_type);
1623 ret->compressed = le32_to_cpu(ret->compressed);
1624 ret->log_offset = le32_to_cpu(ret->log_offset);
1625 ret->log_hist_coarseness = le32_to_cpu(ret->log_hist_coarseness);
1626
1627 if (*store_direct)
1628 return ret;
1629
1630 samples = &ret->samples[0];
1631 for (i = 0; i < ret->nr_samples; i++) {
1632 struct io_sample *s;
1633
1634 s = __get_sample(samples, ret->log_offset, i);
1635 if (ret->log_type == IO_LOG_TYPE_HIST)
1636 s = (struct io_sample *)((char *)s + sizeof(struct io_u_plat_entry) * i);
1637
1638 s->time = le64_to_cpu(s->time);
1639 s->data.val = le64_to_cpu(s->data.val);
1640 s->__ddir = le32_to_cpu(s->__ddir);
1641 s->bs = le32_to_cpu(s->bs);
1642
1643 if (ret->log_offset) {
1644 struct io_sample_offset *so = (void *) s;
1645
1646 so->offset = le64_to_cpu(so->offset);
1647 }
1648
1649 if (ret->log_type == IO_LOG_TYPE_HIST) {
1650 s->data.plat_entry = (struct io_u_plat_entry *)(((char *)s) + sizeof(*s));
1651 s->data.plat_entry->list.next = NULL;
1652 s->data.plat_entry->list.prev = NULL;
1653 }
1654 }
1655
1656 return ret;
1657}
1658
1659static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep,
1660 size_t size, uint64_t tag)
1661{
1662 rep->error = cpu_to_le32(rep->error);
1663 fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL);
1664}
1665
1666static int fio_send_file(struct fio_client *client, struct cmd_sendfile *pdu,
1667 uint64_t tag)
1668{
1669 struct cmd_sendfile_reply *rep;
1670 struct stat sb;
1671 size_t size;
1672 int fd;
1673
1674 size = sizeof(*rep);
1675 rep = malloc(size);
1676
1677 if (stat((char *)pdu->path, &sb) < 0) {
1678fail:
1679 rep->error = errno;
1680 sendfile_reply(client->fd, rep, size, tag);
1681 free(rep);
1682 return 1;
1683 }
1684
1685 size += sb.st_size;
1686 rep = realloc(rep, size);
1687 rep->size = cpu_to_le32((uint32_t) sb.st_size);
1688
1689 fd = open((char *)pdu->path, O_RDONLY);
1690 if (fd == -1 )
1691 goto fail;
1692
1693 rep->error = read_data(fd, &rep->data, sb.st_size);
1694 sendfile_reply(client->fd, rep, size, tag);
1695 free(rep);
1696 close(fd);
1697 return 0;
1698}
1699
1700int fio_handle_client(struct fio_client *client)
1701{
1702 struct client_ops *ops = client->ops;
1703 struct fio_net_cmd *cmd;
1704 int size;
1705
1706 dprint(FD_NET, "client: handle %s\n", client->hostname);
1707
1708 cmd = fio_net_recv_cmd(client->fd, true);
1709 if (!cmd)
1710 return 0;
1711
1712 dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1713 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1714
1715 client->last_cmd = cmd->opcode;
1716
1717 switch (cmd->opcode) {
1718 case FIO_NET_CMD_QUIT:
1719 if (ops->quit)
1720 ops->quit(client, cmd);
1721 remove_client(client);
1722 break;
1723 case FIO_NET_CMD_TEXT:
1724 convert_text(cmd);
1725 ops->text(client, cmd);
1726 break;
1727 case FIO_NET_CMD_DU: {
1728 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1729
1730 convert_dus(&du->dus);
1731 convert_agg(&du->agg);
1732
1733 ops->disk_util(client, cmd);
1734 break;
1735 }
1736 case FIO_NET_CMD_TS: {
1737 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1738
1739 dprint(FD_NET, "client: ts->ss_state = %u\n", (unsigned int) le32_to_cpu(p->ts.ss_state));
1740 if (le32_to_cpu(p->ts.ss_state) & FIO_SS_DATA) {
1741 dprint(FD_NET, "client: received steadystate ring buffers\n");
1742
1743 size = le64_to_cpu(p->ts.ss_dur);
1744 p->ts.ss_iops_data = (uint64_t *) ((struct cmd_ts_pdu *)cmd->payload + 1);
1745 p->ts.ss_bw_data = p->ts.ss_iops_data + size;
1746 }
1747
1748 convert_ts(&p->ts, &p->ts);
1749 convert_gs(&p->rs, &p->rs);
1750
1751 ops->thread_status(client, cmd);
1752 break;
1753 }
1754 case FIO_NET_CMD_GS: {
1755 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1756
1757 convert_gs(gs, gs);
1758
1759 ops->group_stats(client, cmd);
1760 break;
1761 }
1762 case FIO_NET_CMD_ETA: {
1763 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1764
1765 if (!remove_reply_cmd(client, cmd))
1766 break;
1767 convert_jobs_eta(je);
1768 handle_eta(client, cmd);
1769 break;
1770 }
1771 case FIO_NET_CMD_PROBE:
1772 remove_reply_cmd(client, cmd);
1773 ops->probe(client, cmd);
1774 break;
1775 case FIO_NET_CMD_SERVER_START:
1776 client->state = Client_running;
1777 if (ops->job_start)
1778 ops->job_start(client, cmd);
1779 break;
1780 case FIO_NET_CMD_START: {
1781 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1782
1783 pdu->jobs = le32_to_cpu(pdu->jobs);
1784 ops->start(client, cmd);
1785 break;
1786 }
1787 case FIO_NET_CMD_STOP: {
1788 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1789
1790 convert_stop(cmd);
1791 client->state = Client_stopped;
1792 client->error = le32_to_cpu(pdu->error);
1793 client->signal = le32_to_cpu(pdu->signal);
1794 ops->stop(client);
1795 break;
1796 }
1797 case FIO_NET_CMD_ADD_JOB: {
1798 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1799
1800 client->thread_number = le32_to_cpu(pdu->thread_number);
1801 client->groupid = le32_to_cpu(pdu->groupid);
1802
1803 if (ops->add_job)
1804 ops->add_job(client, cmd);
1805 break;
1806 }
1807 case FIO_NET_CMD_IOLOG:
1808 fio_client_handle_iolog(client, cmd);
1809 break;
1810 case FIO_NET_CMD_UPDATE_JOB:
1811 ops->update_job(client, cmd);
1812 remove_reply_cmd(client, cmd);
1813 break;
1814 case FIO_NET_CMD_VTRIGGER: {
1815 struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
1816 char buf[128];
1817 int off = 0;
1818
1819 if (aux_path) {
1820 strcpy(buf, aux_path);
1821 off = strlen(buf);
1822 }
1823
1824 __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
1825 exec_trigger(trigger_cmd);
1826 break;
1827 }
1828 case FIO_NET_CMD_SENDFILE: {
1829 struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload;
1830 fio_send_file(client, pdu, cmd->tag);
1831 break;
1832 }
1833 case FIO_NET_CMD_JOB_OPT: {
1834 handle_job_opt(client, cmd);
1835 break;
1836 }
1837 default:
1838 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1839 break;
1840 }
1841
1842 free(cmd);
1843 return 1;
1844}
1845
1846int fio_clients_send_trigger(const char *cmd)
1847{
1848 struct flist_head *entry;
1849 struct fio_client *client;
1850 size_t slen;
1851
1852 dprint(FD_NET, "client: send vtrigger: %s\n", cmd);
1853
1854 if (!cmd)
1855 slen = 0;
1856 else
1857 slen = strlen(cmd);
1858
1859 flist_for_each(entry, &client_list) {
1860 struct cmd_vtrigger_pdu *pdu;
1861
1862 client = flist_entry(entry, struct fio_client, list);
1863
1864 pdu = malloc(sizeof(*pdu) + slen);
1865 pdu->len = cpu_to_le16((uint16_t) slen);
1866 if (slen)
1867 memcpy(pdu->cmd, cmd, slen);
1868 fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu,
1869 sizeof(*pdu) + slen, NULL, NULL);
1870 free(pdu);
1871 }
1872
1873 return 0;
1874}
1875
1876static void request_client_etas(struct client_ops *ops)
1877{
1878 struct fio_client *client;
1879 struct flist_head *entry;
1880 struct client_eta *eta;
1881 int skipped = 0;
1882
1883 if (eta_print == FIO_ETA_NEVER)
1884 return;
1885
1886 dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1887
1888 eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
1889 eta->pending = nr_clients;
1890
1891 flist_for_each(entry, &client_list) {
1892 client = flist_entry(entry, struct fio_client, list);
1893
1894 if (!flist_empty(&client->eta_list)) {
1895 skipped++;
1896 continue;
1897 }
1898 if (client->state != Client_running)
1899 continue;
1900
1901 assert(!client->eta_in_flight);
1902 flist_add_tail(&client->eta_list, &eta_list);
1903 client->eta_in_flight = eta;
1904 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1905 (uintptr_t) eta, &client->cmd_list);
1906 }
1907
1908 while (skipped--) {
1909 if (!fio_client_dec_jobs_eta(eta, ops->eta))
1910 break;
1911 }
1912
1913 dprint(FD_NET, "client: requested eta tag %p\n", eta);
1914}
1915
1916/*
1917 * A single SEND_ETA timeout isn't fatal. Attempt to recover.
1918 */
1919static int handle_cmd_timeout(struct fio_client *client,
1920 struct fio_net_cmd_reply *reply)
1921{
1922 uint16_t reply_opcode = reply->opcode;
1923
1924 flist_del(&reply->list);
1925 free(reply);
1926
1927 if (reply_opcode != FIO_NET_CMD_SEND_ETA)
1928 return 1;
1929
1930 log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
1931
1932 flist_del_init(&client->eta_list);
1933 if (client->eta_in_flight) {
1934 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
1935 client->eta_in_flight = NULL;
1936 }
1937
1938 /*
1939 * If we fail 5 in a row, give up...
1940 */
1941 if (client->eta_timeouts++ > 5)
1942 return 1;
1943
1944 return 0;
1945}
1946
1947static int client_check_cmd_timeout(struct fio_client *client,
1948 struct timespec *now)
1949{
1950 struct fio_net_cmd_reply *reply;
1951 struct flist_head *entry, *tmp;
1952 int ret = 0;
1953
1954 flist_for_each_safe(entry, tmp, &client->cmd_list) {
1955 unsigned int op;
1956
1957 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1958
1959 if (mtime_since(&reply->ts, now) < FIO_NET_CLIENT_TIMEOUT)
1960 continue;
1961
1962 op = reply->opcode;
1963 if (!handle_cmd_timeout(client, reply))
1964 continue;
1965
1966 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1967 fio_server_op(op));
1968 ret = 1;
1969 }
1970
1971 return flist_empty(&client->cmd_list) && ret;
1972}
1973
1974static int fio_check_clients_timed_out(void)
1975{
1976 struct fio_client *client;
1977 struct flist_head *entry, *tmp;
1978 struct timespec ts;
1979 int ret = 0;
1980
1981 fio_gettime(&ts, NULL);
1982
1983 flist_for_each_safe(entry, tmp, &client_list) {
1984 client = flist_entry(entry, struct fio_client, list);
1985
1986 if (flist_empty(&client->cmd_list))
1987 continue;
1988
1989 if (!client_check_cmd_timeout(client, &ts))
1990 continue;
1991
1992 if (client->ops->timed_out)
1993 client->ops->timed_out(client);
1994 else
1995 log_err("fio: client %s timed out\n", client->hostname);
1996
1997 if (client->last_cmd != FIO_NET_CMD_VTRIGGER)
1998 client->error = ETIMEDOUT;
1999 else
2000 log_info("fio: ignoring timeout due to vtrigger\n");
2001 remove_client(client);
2002 ret = 1;
2003 }
2004
2005 return ret;
2006}
2007
2008int fio_handle_clients(struct client_ops *ops)
2009{
2010 struct pollfd *pfds;
2011 int i, ret = 0, retval = 0;
2012
2013 fio_gettime(&eta_ts, NULL);
2014
2015 pfds = malloc(nr_clients * sizeof(struct pollfd));
2016
2017 init_thread_stat(&client_ts);
2018 init_group_run_stat(&client_gs);
2019
2020 while (!exit_backend && nr_clients) {
2021 struct flist_head *entry, *tmp;
2022 struct fio_client *client;
2023
2024 i = 0;
2025 flist_for_each_safe(entry, tmp, &client_list) {
2026 client = flist_entry(entry, struct fio_client, list);
2027
2028 if (!client->sent_job && !client->ops->stay_connected &&
2029 flist_empty(&client->cmd_list)) {
2030 remove_client(client);
2031 continue;
2032 }
2033
2034 pfds[i].fd = client->fd;
2035 pfds[i].events = POLLIN;
2036 i++;
2037 }
2038
2039 if (!nr_clients)
2040 break;
2041
2042 assert(i == nr_clients);
2043
2044 do {
2045 struct timespec ts;
2046 int timeout;
2047
2048 fio_gettime(&ts, NULL);
2049 if (eta_time_within_slack(mtime_since(&eta_ts, &ts))) {
2050 request_client_etas(ops);
2051 memcpy(&eta_ts, &ts, sizeof(ts));
2052
2053 if (fio_check_clients_timed_out())
2054 break;
2055 }
2056
2057 check_trigger_file();
2058
2059 timeout = min(100u, ops->eta_msec);
2060
2061 ret = poll(pfds, nr_clients, timeout);
2062 if (ret < 0) {
2063 if (errno == EINTR)
2064 continue;
2065 log_err("fio: poll clients: %s\n", strerror(errno));
2066 break;
2067 } else if (!ret)
2068 continue;
2069 } while (ret <= 0);
2070
2071 for (i = 0; i < nr_clients; i++) {
2072 if (!(pfds[i].revents & POLLIN))
2073 continue;
2074
2075 client = find_client_by_fd(pfds[i].fd);
2076 if (!client) {
2077 log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd);
2078 continue;
2079 }
2080 if (!fio_handle_client(client)) {
2081 log_info("client: host=%s disconnected\n",
2082 client->hostname);
2083 remove_client(client);
2084 retval = 1;
2085 } else if (client->error)
2086 retval = 1;
2087 fio_put_client(client);
2088 }
2089 }
2090
2091 fio_client_json_fini();
2092
2093 free(pfds);
2094 return retval || error_clients;
2095}
fio - Flexible IO Tester