9 #include <sys/socket.h>
11 #include <netinet/in.h>
12 #include <arpa/inet.h>
24 #include "verify-state.h"
26 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
27 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd);
28 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd);
29 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
30 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd);
31 static void handle_stop(struct fio_client *client);
32 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd);
34 static void convert_text(struct fio_net_cmd *cmd);
35 static void client_display_thread_status(struct jobs_eta *je);
37 struct client_ops fio_client_ops = {
39 .disk_util = handle_du,
40 .thread_status = handle_ts,
41 .group_stats = handle_gs,
43 .start = handle_start,
44 .eta = client_display_thread_status,
45 .probe = handle_probe,
46 .eta_msec = FIO_CLIENT_DEF_ETA_MSEC,
47 .client_type = FIO_CLIENT_TYPE_CLI,
50 static struct timespec eta_ts;
52 static FLIST_HEAD(client_list);
53 static FLIST_HEAD(eta_list);
55 static FLIST_HEAD(arg_list);
57 struct thread_stat client_ts;
58 struct group_run_stats client_gs;
61 static int sum_stat_nr;
62 static struct buf_output allclients;
63 static struct json_object *root = NULL;
64 static struct json_object *job_opt_object = NULL;
65 static struct json_array *clients_array = NULL;
66 static struct json_array *du_array = NULL;
68 static int error_clients;
70 #define FIO_CLIENT_HASH_BITS 7
71 #define FIO_CLIENT_HASH_SZ (1 << FIO_CLIENT_HASH_BITS)
72 #define FIO_CLIENT_HASH_MASK (FIO_CLIENT_HASH_SZ - 1)
73 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
75 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *, bool *);
77 static void fio_client_add_hash(struct fio_client *client)
79 int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
81 bucket &= FIO_CLIENT_HASH_MASK;
82 flist_add(&client->hash_list, &client_hash[bucket]);
85 static void fio_client_remove_hash(struct fio_client *client)
87 if (!flist_empty(&client->hash_list))
88 flist_del_init(&client->hash_list);
91 static void fio_init fio_client_hash_init(void)
95 for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
96 INIT_FLIST_HEAD(&client_hash[i]);
99 static int read_data(int fd, void *data, size_t size)
104 ret = read(fd, data, size);
106 if (errno == EAGAIN || errno == EINTR)
123 static int read_ini_data(int fd, void *data, size_t size)
134 fp = fdopen(dupfd, "r");
143 char buf[OPT_LEN_MAX+1], *sub;
145 if (!fgets(buf, sizeof(buf), fp)) {
147 if (errno == EAGAIN || errno == EINTR)
154 sub = fio_option_dup_subs(buf);
156 if (len + 1 > size) {
157 log_err("fio: no space left to read data\n");
175 static void fio_client_json_init(void)
180 if (!(output_format & FIO_OUTPUT_JSON))
184 os_ctime_r((const time_t *) &time_p, time_buf, sizeof(time_buf));
185 time_buf[strlen(time_buf) - 1] = '\0';
187 root = json_create_object();
188 json_object_add_value_string(root, "fio version", fio_version_string);
189 json_object_add_value_int(root, "timestamp", time_p);
190 json_object_add_value_string(root, "time", time_buf);
192 job_opt_object = json_create_object();
193 json_object_add_value_object(root, "global options", job_opt_object);
194 clients_array = json_create_array();
195 json_object_add_value_array(root, "client_stats", clients_array);
196 du_array = json_create_array();
197 json_object_add_value_array(root, "disk_util", du_array);
200 static void fio_client_json_fini(void)
202 struct buf_output out;
207 buf_output_init(&out);
209 __log_buf(&out, "\n");
210 json_print_object(root, &out);
211 __log_buf(&out, "\n");
212 log_info_buf(out.buf, out.buflen);
214 buf_output_free(&out);
216 json_free_object(root);
218 job_opt_object = NULL;
219 clients_array = NULL;
223 static struct fio_client *find_client_by_fd(int fd)
225 int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
226 struct fio_client *client;
227 struct flist_head *entry;
229 flist_for_each(entry, &client_hash[bucket]) {
230 client = flist_entry(entry, struct fio_client, hash_list);
232 if (client->fd == fd) {
241 void fio_put_client(struct fio_client *client)
246 log_info_buf(client->buf.buf, client->buf.buflen);
247 buf_output_free(&client->buf);
249 free(client->hostname);
254 while (client->nr_files) {
255 struct client_file *cf = &client->files[--client->nr_files];
261 if (client->opt_lists)
262 free(client->opt_lists);
264 if (!client->did_stat)
273 static int fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
275 if (!--eta->pending) {
284 static void fio_drain_client_text(struct fio_client *client)
287 struct fio_net_cmd *cmd;
289 cmd = fio_net_recv_cmd(client->fd, false);
293 if (cmd->opcode == FIO_NET_CMD_TEXT) {
295 client->ops->text(client, cmd);
302 static void remove_client(struct fio_client *client)
304 assert(client->refs);
306 dprint(FD_NET, "client: removed <%s>\n", client->hostname);
308 fio_drain_client_text(client);
310 if (!flist_empty(&client->list))
311 flist_del_init(&client->list);
313 fio_client_remove_hash(client);
315 if (!flist_empty(&client->eta_list)) {
316 flist_del_init(&client->eta_list);
317 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
323 if (client->ops->removed)
324 client->ops->removed(client);
327 fio_put_client(client);
330 struct fio_client *fio_get_client(struct fio_client *client)
336 static void __fio_client_add_cmd_option(struct fio_client *client,
341 index = client->argc++;
342 client->argv = realloc(client->argv, sizeof(char *) * client->argc);
343 client->argv[index] = strdup(opt);
344 dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
347 void fio_client_add_cmd_option(void *cookie, const char *opt)
349 struct fio_client *client = cookie;
350 struct flist_head *entry;
355 __fio_client_add_cmd_option(client, opt);
358 * Duplicate arguments to shared client group
360 flist_for_each(entry, &arg_list) {
361 client = flist_entry(entry, struct fio_client, arg_list);
363 __fio_client_add_cmd_option(client, opt);
367 static struct fio_client *get_new_client(void)
369 struct fio_client *client;
371 client = malloc(sizeof(*client));
372 memset(client, 0, sizeof(*client));
374 INIT_FLIST_HEAD(&client->list);
375 INIT_FLIST_HEAD(&client->hash_list);
376 INIT_FLIST_HEAD(&client->arg_list);
377 INIT_FLIST_HEAD(&client->eta_list);
378 INIT_FLIST_HEAD(&client->cmd_list);
380 buf_output_init(&client->buf);
385 struct fio_client *fio_client_add_explicit(struct client_ops *ops,
386 const char *hostname, int type,
389 struct fio_client *client;
391 client = get_new_client();
393 client->hostname = strdup(hostname);
395 if (type == Fio_client_socket)
396 client->is_sock = true;
400 ipv6 = type == Fio_client_ipv6;
401 if (fio_server_parse_host(hostname, ipv6,
402 &client->addr.sin_addr,
403 &client->addr6.sin6_addr))
412 client->type = ops->client_type;
414 __fio_client_add_cmd_option(client, "fio");
416 flist_add(&client->list, &client_list);
418 dprint(FD_NET, "client: added <%s>\n", client->hostname);
425 int fio_client_add_ini_file(void *cookie, const char *ini_file, bool remote)
427 struct fio_client *client = cookie;
428 struct client_file *cf;
435 dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file);
437 new_size = (client->nr_files + 1) * sizeof(struct client_file);
438 new_files = realloc(client->files, new_size);
442 client->files = new_files;
443 cf = &client->files[client->nr_files];
444 cf->file = strdup(ini_file);
450 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
452 struct fio_client *existing = *cookie;
453 struct fio_client *client;
457 * We always add our "exec" name as the option, hence 1
460 if (existing->argc == 1)
461 flist_add_tail(&existing->arg_list, &arg_list);
463 while (!flist_empty(&arg_list))
464 flist_del_init(arg_list.next);
468 client = get_new_client();
470 if (fio_server_parse_string(hostname, &client->hostname,
471 &client->is_sock, &client->port,
472 &client->addr.sin_addr,
473 &client->addr6.sin6_addr,
480 client->type = ops->client_type;
482 __fio_client_add_cmd_option(client, "fio");
484 flist_add(&client->list, &client_list);
486 dprint(FD_NET, "client: added <%s>\n", client->hostname);
491 static const char *server_name(struct fio_client *client, char *buf,
497 from = inet_ntop(AF_INET6, (struct sockaddr *) &client->addr6.sin6_addr, buf, bufsize);
498 else if (client->is_sock)
501 from = inet_ntop(AF_INET, (struct sockaddr *) &client->addr.sin_addr, buf, bufsize);
506 static void probe_client(struct fio_client *client)
508 struct cmd_client_probe_pdu pdu;
513 dprint(FD_NET, "client: send probe\n");
516 pdu.flags = __le64_to_cpu(FIO_PROBE_FLAG_ZLIB);
521 sname = server_name(client, buf, sizeof(buf));
522 memset(pdu.server, 0, sizeof(pdu.server));
523 snprintf((char *) pdu.server, sizeof(pdu.server), "%s", sname);
525 fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list);
528 static int fio_client_connect_ip(struct fio_client *client)
530 struct sockaddr *addr;
535 client->addr6.sin6_family = AF_INET6;
536 client->addr6.sin6_port = htons(client->port);
538 addr = (struct sockaddr *) &client->addr6;
539 socklen = sizeof(client->addr6);
541 client->addr.sin_family = AF_INET;
542 client->addr.sin_port = htons(client->port);
544 addr = (struct sockaddr *) &client->addr;
545 socklen = sizeof(client->addr);
548 fd = socket(domain, SOCK_STREAM, 0);
552 log_err("fio: socket: %s\n", strerror(errno));
556 if (connect(fd, addr, socklen) < 0) {
559 log_err("fio: connect: %s\n", strerror(errno));
560 log_err("fio: failed to connect to %s:%u\n", client->hostname,
569 static int fio_client_connect_sock(struct fio_client *client)
571 struct sockaddr_un *addr = &client->addr_un;
575 memset(addr, 0, sizeof(*addr));
576 addr->sun_family = AF_UNIX;
577 snprintf(addr->sun_path, sizeof(addr->sun_path), "%s",
580 fd = socket(AF_UNIX, SOCK_STREAM, 0);
584 log_err("fio: socket: %s\n", strerror(errno));
588 len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
589 if (connect(fd, (struct sockaddr *) addr, len) < 0) {
592 log_err("fio: connect; %s\n", strerror(errno));
600 int fio_client_connect(struct fio_client *client)
604 dprint(FD_NET, "client: connect to host %s\n", client->hostname);
607 fd = fio_client_connect_sock(client);
609 fd = fio_client_connect_ip(client);
611 dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
617 fio_client_add_hash(client);
618 client->state = Client_connected;
620 probe_client(client);
624 int fio_client_terminate(struct fio_client *client)
626 return fio_net_send_quit(client->fd);
629 static void fio_clients_terminate(void)
631 struct flist_head *entry;
632 struct fio_client *client;
634 dprint(FD_NET, "client: terminate clients\n");
636 flist_for_each(entry, &client_list) {
637 client = flist_entry(entry, struct fio_client, list);
638 fio_client_terminate(client);
642 static void sig_int(int sig)
644 dprint(FD_NET, "client: got signal %d\n", sig);
645 fio_clients_terminate();
648 static void client_signal_handler(void)
650 struct sigaction act;
652 memset(&act, 0, sizeof(act));
653 act.sa_handler = sig_int;
654 act.sa_flags = SA_RESTART;
655 sigaction(SIGINT, &act, NULL);
657 memset(&act, 0, sizeof(act));
658 act.sa_handler = sig_int;
659 act.sa_flags = SA_RESTART;
660 sigaction(SIGTERM, &act, NULL);
662 /* Windows uses SIGBREAK as a quit signal from other applications */
664 memset(&act, 0, sizeof(act));
665 act.sa_handler = sig_int;
666 act.sa_flags = SA_RESTART;
667 sigaction(SIGBREAK, &act, NULL);
670 memset(&act, 0, sizeof(act));
671 act.sa_handler = sig_show_status;
672 act.sa_flags = SA_RESTART;
673 sigaction(SIGUSR1, &act, NULL);
676 static int send_client_cmd_line(struct fio_client *client)
678 struct cmd_single_line_pdu *cslp;
679 struct cmd_line_pdu *clp;
680 unsigned long offset;
686 dprint(FD_NET, "client: send cmdline %d\n", client->argc);
688 lens = malloc(client->argc * sizeof(unsigned int));
691 * Find out how much mem we need
693 for (i = 0, mem = 0; i < client->argc; i++) {
694 lens[i] = strlen(client->argv[i]) + 1;
699 * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
701 mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
705 offset = sizeof(*clp);
707 for (i = 0; i < client->argc; i++) {
708 uint16_t arg_len = lens[i];
711 strcpy((char *) cslp->text, client->argv[i]);
712 cslp->len = cpu_to_le16(arg_len);
713 offset += sizeof(*cslp) + arg_len;
717 clp->lines = cpu_to_le16(client->argc);
718 clp->client_type = __cpu_to_le16(client->type);
719 ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, NULL, NULL);
724 int fio_clients_connect(void)
726 struct fio_client *client;
727 struct flist_head *entry, *tmp;
732 WSAStartup(MAKEWORD(2, 2), &wsd);
735 dprint(FD_NET, "client: connect all\n");
737 client_signal_handler();
739 flist_for_each_safe(entry, tmp, &client_list) {
740 client = flist_entry(entry, struct fio_client, list);
742 ret = fio_client_connect(client);
744 remove_client(client);
748 if (client->argc > 1)
749 send_client_cmd_line(client);
755 int fio_start_client(struct fio_client *client)
757 dprint(FD_NET, "client: start %s\n", client->hostname);
758 return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
761 int fio_start_all_clients(void)
763 struct fio_client *client;
764 struct flist_head *entry, *tmp;
767 dprint(FD_NET, "client: start all\n");
769 fio_client_json_init();
771 flist_for_each_safe(entry, tmp, &client_list) {
772 client = flist_entry(entry, struct fio_client, list);
774 ret = fio_start_client(client);
776 remove_client(client);
781 return flist_empty(&client_list);
784 static int __fio_client_send_remote_ini(struct fio_client *client,
785 const char *filename)
787 struct cmd_load_file_pdu *pdu;
791 dprint(FD_NET, "send remote ini %s to %s\n", filename, client->hostname);
793 p_size = sizeof(*pdu) + strlen(filename) + 1;
794 pdu = malloc(p_size);
795 memset(pdu, 0, p_size);
796 pdu->name_len = strlen(filename);
797 strcpy((char *) pdu->file, filename);
798 pdu->client_type = cpu_to_le16((uint16_t) client->type);
800 client->sent_job = true;
801 ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_LOAD_FILE, pdu, p_size,NULL, NULL);
807 * Send file contents to server backend. We could use sendfile(), but to remain
808 * more portable lets just read/write the darn thing.
810 static int __fio_client_send_local_ini(struct fio_client *client,
811 const char *filename)
813 struct cmd_job_pdu *pdu;
821 dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
823 fd = open(filename, O_RDONLY);
826 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
830 if (fstat(fd, &sb) < 0) {
832 log_err("fio: job file stat: %s\n", strerror(errno));
838 * Add extra space for variable expansion, but doesn't guarantee.
840 sb.st_size += OPT_LEN_MAX;
841 p_size = sb.st_size + sizeof(*pdu);
842 pdu = malloc(p_size);
847 if (read_ini_data(fd, p, len)) {
848 log_err("fio: failed reading job file %s\n", filename);
854 pdu->buf_len = __cpu_to_le32(sb.st_size);
855 pdu->client_type = cpu_to_le32(client->type);
857 client->sent_job = true;
858 ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, NULL, NULL);
864 int fio_client_send_ini(struct fio_client *client, const char *filename,
870 ret = __fio_client_send_local_ini(client, filename);
872 ret = __fio_client_send_remote_ini(client, filename);
875 client->sent_job = true;
880 static int fio_client_send_cf(struct fio_client *client,
881 struct client_file *cf)
883 return fio_client_send_ini(client, cf->file, cf->remote);
886 int fio_clients_send_ini(const char *filename)
888 struct fio_client *client;
889 struct flist_head *entry, *tmp;
891 flist_for_each_safe(entry, tmp, &client_list) {
894 client = flist_entry(entry, struct fio_client, list);
896 if (client->nr_files) {
899 for (i = 0; i < client->nr_files; i++) {
900 struct client_file *cf;
902 cf = &client->files[i];
904 if (fio_client_send_cf(client, cf)) {
906 remove_client(client);
911 if (client->sent_job || failed)
913 if (!filename || fio_client_send_ini(client, filename, 0))
914 remove_client(client);
920 int fio_client_update_options(struct fio_client *client,
921 struct thread_options *o, uint64_t *tag)
923 struct cmd_add_job_pdu pdu;
925 pdu.thread_number = cpu_to_le32(client->thread_number);
926 pdu.groupid = cpu_to_le32(client->groupid);
927 convert_thread_options_to_net(&pdu.top, o);
929 return fio_net_send_cmd(client->fd, FIO_NET_CMD_UPDATE_JOB, &pdu, sizeof(pdu), tag, &client->cmd_list);
932 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
934 dst->max_val = le64_to_cpu(src->max_val);
935 dst->min_val = le64_to_cpu(src->min_val);
936 dst->samples = le64_to_cpu(src->samples);
939 * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
941 dst->mean.u.f = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
942 dst->S.u.f = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
945 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
949 dst->error = le32_to_cpu(src->error);
950 dst->thread_number = le32_to_cpu(src->thread_number);
951 dst->groupid = le32_to_cpu(src->groupid);
952 dst->pid = le32_to_cpu(src->pid);
953 dst->members = le32_to_cpu(src->members);
954 dst->unified_rw_rep = le32_to_cpu(src->unified_rw_rep);
956 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
957 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
958 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
959 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
960 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
961 convert_io_stat(&dst->iops_stat[i], &src->iops_stat[i]);
964 dst->usr_time = le64_to_cpu(src->usr_time);
965 dst->sys_time = le64_to_cpu(src->sys_time);
966 dst->ctx = le64_to_cpu(src->ctx);
967 dst->minf = le64_to_cpu(src->minf);
968 dst->majf = le64_to_cpu(src->majf);
969 dst->clat_percentiles = le32_to_cpu(src->clat_percentiles);
970 dst->lat_percentiles = le32_to_cpu(src->lat_percentiles);
971 dst->percentile_precision = le64_to_cpu(src->percentile_precision);
973 for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
974 fio_fp64_t *fps = &src->percentile_list[i];
975 fio_fp64_t *fpd = &dst->percentile_list[i];
977 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
980 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
981 dst->io_u_map[i] = le64_to_cpu(src->io_u_map[i]);
982 dst->io_u_submit[i] = le64_to_cpu(src->io_u_submit[i]);
983 dst->io_u_complete[i] = le64_to_cpu(src->io_u_complete[i]);
986 for (i = 0; i < FIO_IO_U_LAT_N_NR; i++)
987 dst->io_u_lat_n[i] = le64_to_cpu(src->io_u_lat_n[i]);
988 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
989 dst->io_u_lat_u[i] = le64_to_cpu(src->io_u_lat_u[i]);
990 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
991 dst->io_u_lat_m[i] = le64_to_cpu(src->io_u_lat_m[i]);
993 for (i = 0; i < DDIR_RWDIR_CNT; i++)
994 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
995 dst->io_u_plat[i][j] = le64_to_cpu(src->io_u_plat[i][j]);
997 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
998 dst->total_io_u[i] = le64_to_cpu(src->total_io_u[i]);
999 dst->short_io_u[i] = le64_to_cpu(src->short_io_u[i]);
1000 dst->drop_io_u[i] = le64_to_cpu(src->drop_io_u[i]);
1003 dst->total_submit = le64_to_cpu(src->total_submit);
1004 dst->total_complete = le64_to_cpu(src->total_complete);
1005 dst->nr_zone_resets = le64_to_cpu(src->nr_zone_resets);
1007 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1008 dst->io_bytes[i] = le64_to_cpu(src->io_bytes[i]);
1009 dst->runtime[i] = le64_to_cpu(src->runtime[i]);
1012 dst->total_run_time = le64_to_cpu(src->total_run_time);
1013 dst->continue_on_error = le16_to_cpu(src->continue_on_error);
1014 dst->total_err_count = le64_to_cpu(src->total_err_count);
1015 dst->first_error = le32_to_cpu(src->first_error);
1016 dst->kb_base = le32_to_cpu(src->kb_base);
1017 dst->unit_base = le32_to_cpu(src->unit_base);
1019 dst->sig_figs = le32_to_cpu(src->sig_figs);
1021 dst->latency_depth = le32_to_cpu(src->latency_depth);
1022 dst->latency_target = le64_to_cpu(src->latency_target);
1023 dst->latency_window = le64_to_cpu(src->latency_window);
1024 dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
1026 dst->nr_block_infos = le64_to_cpu(src->nr_block_infos);
1027 for (i = 0; i < dst->nr_block_infos; i++)
1028 dst->block_infos[i] = le32_to_cpu(src->block_infos[i]);
1030 dst->ss_dur = le64_to_cpu(src->ss_dur);
1031 dst->ss_state = le32_to_cpu(src->ss_state);
1032 dst->ss_head = le32_to_cpu(src->ss_head);
1033 dst->ss_limit.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_limit.u.i));
1034 dst->ss_slope.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_slope.u.i));
1035 dst->ss_deviation.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_deviation.u.i));
1036 dst->ss_criterion.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_criterion.u.i));
1038 if (dst->ss_state & FIO_SS_DATA) {
1039 for (i = 0; i < dst->ss_dur; i++ ) {
1040 dst->ss_iops_data[i] = le64_to_cpu(src->ss_iops_data[i]);
1041 dst->ss_bw_data[i] = le64_to_cpu(src->ss_bw_data[i]);
1045 dst->cachehit = le64_to_cpu(src->cachehit);
1046 dst->cachemiss = le64_to_cpu(src->cachemiss);
1049 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
1053 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1054 dst->max_run[i] = le64_to_cpu(src->max_run[i]);
1055 dst->min_run[i] = le64_to_cpu(src->min_run[i]);
1056 dst->max_bw[i] = le64_to_cpu(src->max_bw[i]);
1057 dst->min_bw[i] = le64_to_cpu(src->min_bw[i]);
1058 dst->iobytes[i] = le64_to_cpu(src->iobytes[i]);
1059 dst->agg[i] = le64_to_cpu(src->agg[i]);
1062 dst->kb_base = le32_to_cpu(src->kb_base);
1063 dst->unit_base = le32_to_cpu(src->unit_base);
1064 dst->sig_figs = le32_to_cpu(src->sig_figs);
1065 dst->groupid = le32_to_cpu(src->groupid);
1066 dst->unified_rw_rep = le32_to_cpu(src->unified_rw_rep);
1069 static void json_object_add_client_info(struct json_object *obj,
1070 struct fio_client *client)
1072 const char *hostname = client->hostname ? client->hostname : "";
1074 json_object_add_value_string(obj, "hostname", hostname);
1075 json_object_add_value_int(obj, "port", client->port);
1078 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
1080 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1081 struct flist_head *opt_list = NULL;
1082 struct json_object *tsobj;
1084 if (client->opt_lists && p->ts.thread_number <= client->jobs)
1085 opt_list = &client->opt_lists[p->ts.thread_number - 1];
1087 tsobj = show_thread_status(&p->ts, &p->rs, opt_list, &client->buf);
1088 client->did_stat = true;
1090 json_object_add_client_info(tsobj, client);
1091 json_array_add_value_object(clients_array, tsobj);
1094 if (sum_stat_clients <= 1)
1097 sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1);
1098 sum_group_stats(&client_gs, &p->rs);
1100 client_ts.members++;
1101 client_ts.thread_number = p->ts.thread_number;
1102 client_ts.groupid = p->ts.groupid;
1103 client_ts.unified_rw_rep = p->ts.unified_rw_rep;
1104 client_ts.sig_figs = p->ts.sig_figs;
1106 if (++sum_stat_nr == sum_stat_clients) {
1107 strcpy(client_ts.name, "All clients");
1108 tsobj = show_thread_status(&client_ts, &client_gs, NULL, &allclients);
1110 json_object_add_client_info(tsobj, client);
1111 json_array_add_value_object(clients_array, tsobj);
1116 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
1118 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1120 if (output_format & FIO_OUTPUT_NORMAL)
1121 show_group_stats(gs, &client->buf);
1124 static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
1126 struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
1127 struct print_option *p;
1129 if (!job_opt_object)
1132 pdu->global = le16_to_cpu(pdu->global);
1133 pdu->truncated = le16_to_cpu(pdu->truncated);
1134 pdu->groupid = le32_to_cpu(pdu->groupid);
1136 p = malloc(sizeof(*p));
1137 p->name = strdup((char *) pdu->name);
1138 if (pdu->value[0] != '\0')
1139 p->value = strdup((char *) pdu->value);
1144 const char *pos = "";
1149 json_object_add_value_string(job_opt_object, p->name, pos);
1150 } else if (client->opt_lists) {
1151 struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
1153 flist_add_tail(&p->list, opt_list);
1157 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
1159 struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1160 const char *buf = (const char *) pdu->buf;
1163 struct buf_output out;
1165 buf_output_init(&out);
1167 name = client->name ? client->name : client->hostname;
1169 if (!client->skip_newline && !(output_format & FIO_OUTPUT_TERSE))
1170 __log_buf(&out, "<%s> ", name);
1171 __log_buf(&out, "%s", buf);
1172 log_info_buf(out.buf, out.buflen);
1173 buf_output_free(&out);
1174 client->skip_newline = strchr(buf, '\n') == NULL;
1177 static void convert_agg(struct disk_util_agg *agg)
1181 for (i = 0; i < 2; i++) {
1182 agg->ios[i] = le64_to_cpu(agg->ios[i]);
1183 agg->merges[i] = le64_to_cpu(agg->merges[i]);
1184 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
1185 agg->ticks[i] = le64_to_cpu(agg->ticks[i]);
1188 agg->io_ticks = le64_to_cpu(agg->io_ticks);
1189 agg->time_in_queue = le64_to_cpu(agg->time_in_queue);
1190 agg->slavecount = le32_to_cpu(agg->slavecount);
1191 agg->max_util.u.f = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
1194 static void convert_dus(struct disk_util_stat *dus)
1198 for (i = 0; i < 2; i++) {
1199 dus->s.ios[i] = le64_to_cpu(dus->s.ios[i]);
1200 dus->s.merges[i] = le64_to_cpu(dus->s.merges[i]);
1201 dus->s.sectors[i] = le64_to_cpu(dus->s.sectors[i]);
1202 dus->s.ticks[i] = le64_to_cpu(dus->s.ticks[i]);
1205 dus->s.io_ticks = le64_to_cpu(dus->s.io_ticks);
1206 dus->s.time_in_queue = le64_to_cpu(dus->s.time_in_queue);
1207 dus->s.msec = le64_to_cpu(dus->s.msec);
1210 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
1212 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1214 if (!client->disk_stats_shown)
1215 client->disk_stats_shown = true;
1217 if (output_format & FIO_OUTPUT_JSON) {
1218 struct json_object *duobj;
1220 json_array_add_disk_util(&du->dus, &du->agg, du_array);
1221 duobj = json_array_last_value_object(du_array);
1222 json_object_add_client_info(duobj, client);
1224 if (output_format & FIO_OUTPUT_NORMAL) {
1225 __log_buf(&client->buf, "\nDisk stats (read/write):\n");
1226 print_disk_util(&du->dus, &du->agg, 0, &client->buf);
1228 if (output_format & FIO_OUTPUT_TERSE && terse_version >= 3) {
1229 print_disk_util(&du->dus, &du->agg, 1, &client->buf);
1230 __log_buf(&client->buf, "\n");
1234 static void convert_jobs_eta(struct jobs_eta *je)
1238 je->nr_running = le32_to_cpu(je->nr_running);
1239 je->nr_ramp = le32_to_cpu(je->nr_ramp);
1240 je->nr_pending = le32_to_cpu(je->nr_pending);
1241 je->nr_setting_up = le32_to_cpu(je->nr_setting_up);
1242 je->files_open = le32_to_cpu(je->files_open);
1244 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1245 je->m_rate[i] = le64_to_cpu(je->m_rate[i]);
1246 je->t_rate[i] = le64_to_cpu(je->t_rate[i]);
1247 je->m_iops[i] = le32_to_cpu(je->m_iops[i]);
1248 je->t_iops[i] = le32_to_cpu(je->t_iops[i]);
1249 je->rate[i] = le64_to_cpu(je->rate[i]);
1250 je->iops[i] = le32_to_cpu(je->iops[i]);
1253 je->elapsed_sec = le64_to_cpu(je->elapsed_sec);
1254 je->eta_sec = le64_to_cpu(je->eta_sec);
1255 je->nr_threads = le32_to_cpu(je->nr_threads);
1256 je->is_pow2 = le32_to_cpu(je->is_pow2);
1257 je->unit_base = le32_to_cpu(je->unit_base);
1258 je->sig_figs = le32_to_cpu(je->sig_figs);
1261 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
1265 dst->nr_running += je->nr_running;
1266 dst->nr_ramp += je->nr_ramp;
1267 dst->nr_pending += je->nr_pending;
1268 dst->nr_setting_up += je->nr_setting_up;
1269 dst->files_open += je->files_open;
1271 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1272 dst->m_rate[i] += je->m_rate[i];
1273 dst->t_rate[i] += je->t_rate[i];
1274 dst->m_iops[i] += je->m_iops[i];
1275 dst->t_iops[i] += je->t_iops[i];
1276 dst->rate[i] += je->rate[i];
1277 dst->iops[i] += je->iops[i];
1280 dst->elapsed_sec += je->elapsed_sec;
1282 if (je->eta_sec > dst->eta_sec)
1283 dst->eta_sec = je->eta_sec;
1285 dst->nr_threads += je->nr_threads;
1288 * This wont be correct for multiple strings, but at least it
1289 * works for the basic cases.
1291 strcpy((char *) dst->run_str, (char *) je->run_str);
1294 static bool remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
1296 struct fio_net_cmd_reply *reply = NULL;
1297 struct flist_head *entry;
1299 flist_for_each(entry, &client->cmd_list) {
1300 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1302 if (cmd->tag == (uintptr_t) reply)
1309 log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag);
1313 flist_del(&reply->list);
1314 cmd->tag = reply->saved_tag;
1319 int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
1322 struct fio_net_cmd_reply *reply = NULL;
1323 struct flist_head *entry;
1325 flist_for_each(entry, &client->cmd_list) {
1326 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1328 if (tag == (uintptr_t) reply)
1343 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
1345 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1346 struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
1348 dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
1350 assert(client->eta_in_flight == eta);
1352 client->eta_in_flight = NULL;
1353 flist_del_init(&client->eta_list);
1354 client->eta_timeouts = 0;
1356 if (client->ops->jobs_eta)
1357 client->ops->jobs_eta(client, je);
1359 fio_client_sum_jobs_eta(&eta->eta, je);
1360 fio_client_dec_jobs_eta(eta, client->ops->eta);
1363 static void client_flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
1364 uint64_t sample_size)
1366 struct io_sample *s;
1368 uint64_t i, j, nr_samples;
1369 struct io_u_plat_entry *entry;
1370 uint64_t *io_u_plat;
1372 int stride = 1 << hist_coarseness;
1377 s = __get_sample(samples, 0, 0);
1378 log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
1380 nr_samples = sample_size / __log_entry_sz(log_offset);
1382 for (i = 0; i < nr_samples; i++) {
1384 s = (struct io_sample *)((char *)__get_sample(samples, log_offset, i) +
1385 i * sizeof(struct io_u_plat_entry));
1387 entry = s->data.plat_entry;
1388 io_u_plat = entry->io_u_plat;
1390 fprintf(f, "%lu, %u, %llu, ", (unsigned long) s->time,
1391 io_sample_ddir(s), (unsigned long long) s->bs);
1392 for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) {
1393 fprintf(f, "%llu, ", (unsigned long long)hist_sum(j, stride, io_u_plat, NULL));
1395 fprintf(f, "%llu\n", (unsigned long long)
1396 hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat, NULL));
1401 static int fio_client_handle_iolog(struct fio_client *client,
1402 struct fio_net_cmd *cmd)
1404 struct cmd_iolog_pdu *pdu = NULL;
1406 char *log_pathname = NULL;
1409 pdu = convert_iolog(cmd, &store_direct);
1411 log_err("fio: failed converting IO log\n");
1416 /* allocate buffer big enough for next sprintf() call */
1417 log_pathname = malloc(10 + strlen((char *)pdu->name) +
1418 strlen(client->hostname));
1419 if (!log_pathname) {
1420 log_err("fio: memory allocation of unique pathname failed\n");
1424 /* generate a unique pathname for the log file using hostname */
1425 sprintf(log_pathname, "%s.%s", pdu->name, client->hostname);
1432 fd = open((const char *) log_pathname,
1433 O_WRONLY | O_CREAT | O_TRUNC, 0644);
1435 log_err("fio: open log %s: %s\n",
1436 log_pathname, strerror(errno));
1441 sz = cmd->pdu_len - sizeof(*pdu);
1442 wrote = write(fd, pdu->samples, sz);
1446 log_err("fio: short write on compressed log\n");
1454 f = fopen((const char *) log_pathname, "w");
1456 log_err("fio: fopen log %s : %s\n",
1457 log_pathname, strerror(errno));
1462 if (pdu->log_type == IO_LOG_TYPE_HIST) {
1463 client_flush_hist_samples(f, pdu->log_hist_coarseness, pdu->samples,
1464 pdu->nr_samples * sizeof(struct io_sample));
1466 flush_samples(f, pdu->samples,
1467 pdu->nr_samples * sizeof(struct io_sample));
1474 if (pdu && pdu != (void *) cmd->payload)
1483 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
1485 struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
1486 const char *os, *arch;
1489 os = fio_get_os_string(probe->os);
1493 arch = fio_get_arch_string(probe->arch);
1497 sprintf(bit, "%d-bit", probe->bpp * 8);
1498 probe->flags = le64_to_cpu(probe->flags);
1500 if (output_format & FIO_OUTPUT_NORMAL) {
1501 log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n",
1502 probe->hostname, probe->bigendian, bit, os, arch,
1503 probe->fio_version, (unsigned long) probe->flags);
1507 client->name = strdup((char *) probe->hostname);
1510 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
1512 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1514 client->state = Client_started;
1515 client->jobs = le32_to_cpu(pdu->jobs);
1516 client->nr_stat = le32_to_cpu(pdu->stat_outputs);
1521 if (client->opt_lists)
1522 free(client->opt_lists);
1524 client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
1525 for (i = 0; i < client->jobs; i++)
1526 INIT_FLIST_HEAD(&client->opt_lists[i]);
1529 sum_stat_clients += client->nr_stat;
1532 static void handle_stop(struct fio_client *client)
1535 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
1538 static void convert_stop(struct fio_net_cmd *cmd)
1540 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1542 pdu->error = le32_to_cpu(pdu->error);
1545 static void convert_text(struct fio_net_cmd *cmd)
1547 struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1549 pdu->level = le32_to_cpu(pdu->level);
1550 pdu->buf_len = le32_to_cpu(pdu->buf_len);
1551 pdu->log_sec = le64_to_cpu(pdu->log_sec);
1552 pdu->log_usec = le64_to_cpu(pdu->log_usec);
1555 static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
1556 struct cmd_iolog_pdu *pdu)
1559 struct cmd_iolog_pdu *ret;
1561 uint64_t nr_samples;
1565 stream.zalloc = Z_NULL;
1566 stream.zfree = Z_NULL;
1567 stream.opaque = Z_NULL;
1568 stream.avail_in = 0;
1569 stream.next_in = Z_NULL;
1571 if (inflateInit(&stream) != Z_OK)
1575 * Get header first, it's not compressed
1577 nr_samples = le64_to_cpu(pdu->nr_samples);
1579 if (pdu->log_type == IO_LOG_TYPE_HIST)
1580 total = nr_samples * (__log_entry_sz(le32_to_cpu(pdu->log_offset)) +
1581 sizeof(struct io_u_plat_entry));
1583 total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
1584 ret = malloc(total + sizeof(*pdu));
1585 ret->nr_samples = nr_samples;
1587 memcpy(ret, pdu, sizeof(*pdu));
1589 p = (char *) ret + sizeof(*pdu);
1591 stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1592 stream.next_in = (void *)((char *) pdu + sizeof(*pdu));
1593 while (stream.avail_in) {
1594 unsigned int this_chunk = 65536;
1595 unsigned int this_len;
1598 if (this_chunk > total)
1601 stream.avail_out = this_chunk;
1602 stream.next_out = (void *)p;
1603 err = inflate(&stream, Z_NO_FLUSH);
1604 /* may be Z_OK, or Z_STREAM_END */
1607 * Z_STREAM_ERROR and Z_BUF_ERROR can safely be
1609 if (err == Z_STREAM_ERROR || err == Z_BUF_ERROR)
1611 log_err("fio: inflate error %d\n", err);
1617 this_len = this_chunk - stream.avail_out;
1623 inflateEnd(&stream);
1631 * This has been compressed on the server side, since it can be big.
1634 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd,
1637 struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1638 struct cmd_iolog_pdu *ret;
1643 *store_direct = false;
1646 * Convert if compressed and we support it. If it's not
1647 * compressed, we need not do anything.
1649 compressed = le32_to_cpu(pdu->compressed);
1650 if (compressed == XMIT_COMPRESSED) {
1652 log_err("fio: server sent compressed data by mistake\n");
1655 ret = convert_iolog_gz(cmd, pdu);
1657 log_err("fio: failed decompressing log\n");
1660 } else if (compressed == STORE_COMPRESSED) {
1661 *store_direct = true;
1666 ret->nr_samples = le64_to_cpu(ret->nr_samples);
1667 ret->thread_number = le32_to_cpu(ret->thread_number);
1668 ret->log_type = le32_to_cpu(ret->log_type);
1669 ret->compressed = le32_to_cpu(ret->compressed);
1670 ret->log_offset = le32_to_cpu(ret->log_offset);
1671 ret->log_hist_coarseness = le32_to_cpu(ret->log_hist_coarseness);
1676 samples = &ret->samples[0];
1677 for (i = 0; i < ret->nr_samples; i++) {
1678 struct io_sample *s;
1680 s = __get_sample(samples, ret->log_offset, i);
1681 if (ret->log_type == IO_LOG_TYPE_HIST)
1682 s = (struct io_sample *)((char *)s + sizeof(struct io_u_plat_entry) * i);
1684 s->time = le64_to_cpu(s->time);
1685 s->data.val = le64_to_cpu(s->data.val);
1686 s->__ddir = le32_to_cpu(s->__ddir);
1687 s->bs = le64_to_cpu(s->bs);
1689 if (ret->log_offset) {
1690 struct io_sample_offset *so = (void *) s;
1692 so->offset = le64_to_cpu(so->offset);
1695 if (ret->log_type == IO_LOG_TYPE_HIST) {
1696 s->data.plat_entry = (struct io_u_plat_entry *)(((char *)s) + sizeof(*s));
1697 s->data.plat_entry->list.next = NULL;
1698 s->data.plat_entry->list.prev = NULL;
1705 static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep,
1706 size_t size, uint64_t tag)
1708 rep->error = cpu_to_le32(rep->error);
1709 fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL);
1712 static int fio_send_file(struct fio_client *client, struct cmd_sendfile *pdu,
1715 struct cmd_sendfile_reply *rep;
1720 size = sizeof(*rep);
1723 if (stat((char *)pdu->path, &sb) < 0) {
1726 sendfile_reply(client->fd, rep, size, tag);
1732 rep = realloc(rep, size);
1733 rep->size = cpu_to_le32((uint32_t) sb.st_size);
1735 fd = open((char *)pdu->path, O_RDONLY);
1739 rep->error = read_data(fd, &rep->data, sb.st_size);
1740 sendfile_reply(client->fd, rep, size, tag);
1746 int fio_handle_client(struct fio_client *client)
1748 struct client_ops *ops = client->ops;
1749 struct fio_net_cmd *cmd;
1752 dprint(FD_NET, "client: handle %s\n", client->hostname);
1754 cmd = fio_net_recv_cmd(client->fd, true);
1758 dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1759 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1761 client->last_cmd = cmd->opcode;
1763 switch (cmd->opcode) {
1764 case FIO_NET_CMD_QUIT:
1766 ops->quit(client, cmd);
1767 remove_client(client);
1769 case FIO_NET_CMD_TEXT:
1771 ops->text(client, cmd);
1773 case FIO_NET_CMD_DU: {
1774 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1776 convert_dus(&du->dus);
1777 convert_agg(&du->agg);
1779 ops->disk_util(client, cmd);
1782 case FIO_NET_CMD_TS: {
1783 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1785 dprint(FD_NET, "client: ts->ss_state = %u\n", (unsigned int) le32_to_cpu(p->ts.ss_state));
1786 if (le32_to_cpu(p->ts.ss_state) & FIO_SS_DATA) {
1787 dprint(FD_NET, "client: received steadystate ring buffers\n");
1789 size = le64_to_cpu(p->ts.ss_dur);
1790 p->ts.ss_iops_data = (uint64_t *) ((struct cmd_ts_pdu *)cmd->payload + 1);
1791 p->ts.ss_bw_data = p->ts.ss_iops_data + size;
1794 convert_ts(&p->ts, &p->ts);
1795 convert_gs(&p->rs, &p->rs);
1797 ops->thread_status(client, cmd);
1800 case FIO_NET_CMD_GS: {
1801 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1805 ops->group_stats(client, cmd);
1808 case FIO_NET_CMD_ETA: {
1809 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1811 if (!remove_reply_cmd(client, cmd))
1813 convert_jobs_eta(je);
1814 handle_eta(client, cmd);
1817 case FIO_NET_CMD_PROBE:
1818 remove_reply_cmd(client, cmd);
1819 ops->probe(client, cmd);
1821 case FIO_NET_CMD_SERVER_START:
1822 client->state = Client_running;
1824 ops->job_start(client, cmd);
1826 case FIO_NET_CMD_START: {
1827 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1829 pdu->jobs = le32_to_cpu(pdu->jobs);
1830 ops->start(client, cmd);
1833 case FIO_NET_CMD_STOP: {
1834 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1837 client->state = Client_stopped;
1838 client->error = le32_to_cpu(pdu->error);
1839 client->signal = le32_to_cpu(pdu->signal);
1843 case FIO_NET_CMD_ADD_JOB: {
1844 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1846 client->thread_number = le32_to_cpu(pdu->thread_number);
1847 client->groupid = le32_to_cpu(pdu->groupid);
1850 ops->add_job(client, cmd);
1853 case FIO_NET_CMD_IOLOG:
1854 fio_client_handle_iolog(client, cmd);
1856 case FIO_NET_CMD_UPDATE_JOB:
1857 ops->update_job(client, cmd);
1858 remove_reply_cmd(client, cmd);
1860 case FIO_NET_CMD_VTRIGGER: {
1861 struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
1866 strcpy(buf, aux_path);
1870 __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
1871 exec_trigger(trigger_cmd);
1874 case FIO_NET_CMD_SENDFILE: {
1875 struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload;
1876 fio_send_file(client, pdu, cmd->tag);
1879 case FIO_NET_CMD_JOB_OPT: {
1880 handle_job_opt(client, cmd);
1884 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1892 int fio_clients_send_trigger(const char *cmd)
1894 struct flist_head *entry;
1895 struct fio_client *client;
1898 dprint(FD_NET, "client: send vtrigger: %s\n", cmd);
1905 flist_for_each(entry, &client_list) {
1906 struct cmd_vtrigger_pdu *pdu;
1908 client = flist_entry(entry, struct fio_client, list);
1910 pdu = malloc(sizeof(*pdu) + slen);
1911 pdu->len = cpu_to_le16((uint16_t) slen);
1913 memcpy(pdu->cmd, cmd, slen);
1914 fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu,
1915 sizeof(*pdu) + slen, NULL, NULL);
1922 static void request_client_etas(struct client_ops *ops)
1924 struct fio_client *client;
1925 struct flist_head *entry;
1926 struct client_eta *eta;
1929 if (eta_print == FIO_ETA_NEVER)
1932 dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1934 eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
1935 eta->pending = nr_clients;
1937 flist_for_each(entry, &client_list) {
1938 client = flist_entry(entry, struct fio_client, list);
1940 if (!flist_empty(&client->eta_list)) {
1944 if (client->state != Client_running)
1947 assert(!client->eta_in_flight);
1948 flist_add_tail(&client->eta_list, &eta_list);
1949 client->eta_in_flight = eta;
1950 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1951 (uintptr_t) eta, &client->cmd_list);
1955 if (!fio_client_dec_jobs_eta(eta, ops->eta))
1959 dprint(FD_NET, "client: requested eta tag %p\n", eta);
1963 * A single SEND_ETA timeout isn't fatal. Attempt to recover.
1965 static int handle_cmd_timeout(struct fio_client *client,
1966 struct fio_net_cmd_reply *reply)
1968 uint16_t reply_opcode = reply->opcode;
1970 flist_del(&reply->list);
1973 if (reply_opcode != FIO_NET_CMD_SEND_ETA)
1976 log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
1978 flist_del_init(&client->eta_list);
1979 if (client->eta_in_flight) {
1980 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
1981 client->eta_in_flight = NULL;
1985 * If we fail 5 in a row, give up...
1987 if (client->eta_timeouts++ > 5)
1993 static int client_check_cmd_timeout(struct fio_client *client,
1994 struct timespec *now)
1996 struct fio_net_cmd_reply *reply;
1997 struct flist_head *entry, *tmp;
2000 flist_for_each_safe(entry, tmp, &client->cmd_list) {
2003 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
2005 if (mtime_since(&reply->ts, now) < FIO_NET_CLIENT_TIMEOUT)
2009 if (!handle_cmd_timeout(client, reply))
2012 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
2017 return flist_empty(&client->cmd_list) && ret;
2020 static int fio_check_clients_timed_out(void)
2022 struct fio_client *client;
2023 struct flist_head *entry, *tmp;
2027 fio_gettime(&ts, NULL);
2029 flist_for_each_safe(entry, tmp, &client_list) {
2030 client = flist_entry(entry, struct fio_client, list);
2032 if (flist_empty(&client->cmd_list))
2035 if (!client_check_cmd_timeout(client, &ts))
2038 if (client->ops->timed_out)
2039 client->ops->timed_out(client);
2041 log_err("fio: client %s timed out\n", client->hostname);
2043 if (client->last_cmd != FIO_NET_CMD_VTRIGGER)
2044 client->error = ETIMEDOUT;
2046 log_info("fio: ignoring timeout due to vtrigger\n");
2047 remove_client(client);
2054 int fio_handle_clients(struct client_ops *ops)
2056 struct pollfd *pfds;
2057 int i, ret = 0, retval = 0;
2059 fio_gettime(&eta_ts, NULL);
2061 pfds = malloc(nr_clients * sizeof(struct pollfd));
2063 init_thread_stat(&client_ts);
2064 init_group_run_stat(&client_gs);
2066 while (!exit_backend && nr_clients) {
2067 struct flist_head *entry, *tmp;
2068 struct fio_client *client;
2071 flist_for_each_safe(entry, tmp, &client_list) {
2072 client = flist_entry(entry, struct fio_client, list);
2074 if (!client->sent_job && !client->ops->stay_connected &&
2075 flist_empty(&client->cmd_list)) {
2076 remove_client(client);
2080 pfds[i].fd = client->fd;
2081 pfds[i].events = POLLIN;
2088 assert(i == nr_clients);
2094 fio_gettime(&ts, NULL);
2095 if (eta_time_within_slack(mtime_since(&eta_ts, &ts))) {
2096 request_client_etas(ops);
2097 memcpy(&eta_ts, &ts, sizeof(ts));
2099 if (fio_check_clients_timed_out())
2103 check_trigger_file();
2105 timeout = min(100u, ops->eta_msec);
2107 ret = poll(pfds, nr_clients, timeout);
2111 log_err("fio: poll clients: %s\n", strerror(errno));
2117 for (i = 0; i < nr_clients; i++) {
2118 if (!(pfds[i].revents & POLLIN))
2121 client = find_client_by_fd(pfds[i].fd);
2123 log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd);
2126 if (!fio_handle_client(client)) {
2127 log_info("client: host=%s disconnected\n",
2129 remove_client(client);
2131 } else if (client->error)
2133 fio_put_client(client);
2137 log_info_buf(allclients.buf, allclients.buflen);
2138 buf_output_free(&allclients);
2140 fio_client_json_fini();
2143 return retval || error_clients;
2146 static void client_display_thread_status(struct jobs_eta *je)
2148 if (!(output_format & FIO_OUTPUT_JSON))
2149 display_thread_status(je);