client.c

   1 #include <stdio.h>
   2 #include <stdlib.h>
   3 #include <unistd.h>
   4 #include <limits.h>
   5 #include <errno.h>
   6 #include <fcntl.h>
   7 #include <sys/poll.h>
   8 #include <sys/types.h>
   9 #include <sys/stat.h>
  10 #include <sys/wait.h>
  11 #include <sys/socket.h>
  12 #include <sys/un.h>
  13 #include <netinet/in.h>
  14 #include <arpa/inet.h>
  15 #include <netdb.h>
  16 #include <signal.h>
  17
  18 #include "fio.h"
  19 #include "client.h"
  20 #include "server.h"
  21 #include "flist.h"
  22 #include "hash.h"
  23
  24 static void fio_client_text_op(struct fio_client *client,
  25                 FILE *f, __u16 pdu_len, const char *buf)
  26 {
  27         const char *name;
  28         int fio_unused ret;
  29
  30         name = client->name ? client->name : client->hostname;
  31
  32         if (!client->skip_newline)
  33                 fprintf(f, "<%s> ", name);
  34         ret = fwrite(buf, pdu_len, 1, f);
  35         fflush(f);
  36         client->skip_newline = strchr(buf, '\n') == NULL;
  37 }
  38
  39 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
  40 static void handle_ts(struct fio_net_cmd *cmd);
  41 static void handle_gs(struct fio_net_cmd *cmd);
  42 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd);
  43 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
  44
  45 struct client_ops fio_client_ops = {
  46         .text_op        = fio_client_text_op,
  47         .disk_util      = handle_du,
  48         .thread_status  = handle_ts,
  49         .group_stats    = handle_gs,
  50         .eta            = handle_eta,
  51         .probe          = handle_probe,
  52 };
  53
  54 static struct timeval eta_tv;
  55
  56 enum {
  57         Client_created          = 0,
  58         Client_connected        = 1,
  59         Client_started          = 2,
  60         Client_running          = 3,
  61         Client_stopped          = 4,
  62         Client_exited           = 5,
  63 };
  64
  65 static FLIST_HEAD(client_list);
  66 static FLIST_HEAD(eta_list);
  67
  68 static FLIST_HEAD(arg_list);
  69
  70 static struct thread_stat client_ts;
  71 static struct group_run_stats client_gs;
  72 static int sum_stat_clients;
  73 static int sum_stat_nr;
  74
  75 #define FIO_CLIENT_HASH_BITS    7
  76 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  77 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  78 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  79
  80 static void fio_client_add_hash(struct fio_client *client)
  81 {
  82         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  83
  84         bucket &= FIO_CLIENT_HASH_MASK;
  85         flist_add(&client->hash_list, &client_hash[bucket]);
  86 }
  87
  88 static void fio_client_remove_hash(struct fio_client *client)
  89 {
  90         if (!flist_empty(&client->hash_list))
  91                 flist_del_init(&client->hash_list);
  92 }
  93
  94 static void fio_init fio_client_hash_init(void)
  95 {
  96         int i;
  97
  98         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
  99                 INIT_FLIST_HEAD(&client_hash[i]);
 100 }
 101
 102 static struct fio_client *find_client_by_fd(int fd)
 103 {
 104         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
 105         struct fio_client *client;
 106         struct flist_head *entry;
 107
 108         flist_for_each(entry, &client_hash[bucket]) {
 109                 client = flist_entry(entry, struct fio_client, hash_list);
 110
 111                 if (client->fd == fd)
 112                         return client;
 113         }
 114
 115         return NULL;
 116 }
 117
 118 static void remove_client(struct fio_client *client)
 119 {
 120         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 121         flist_del(&client->list);
 122
 123         fio_client_remove_hash(client);
 124
 125         if (!flist_empty(&client->eta_list)) {
 126                 flist_del_init(&client->eta_list);
 127                 fio_client_dec_jobs_eta(client->eta_in_flight, display_thread_status);
 128         }
 129
 130         free(client->hostname);
 131         if (client->argv)
 132                 free(client->argv);
 133         if (client->name)
 134                 free(client->name);
 135
 136         free(client);
 137         nr_clients--;
 138         sum_stat_clients--;
 139 }
 140
 141 static void __fio_client_add_cmd_option(struct fio_client *client,
 142                                         const char *opt)
 143 {
 144         int index;
 145
 146         index = client->argc++;
 147         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 148         client->argv[index] = strdup(opt);
 149         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 150 }
 151
 152 void fio_client_add_cmd_option(void *cookie, const char *opt)
 153 {
 154         struct fio_client *client = cookie;
 155         struct flist_head *entry;
 156
 157         if (!client || !opt)
 158                 return;
 159
 160         __fio_client_add_cmd_option(client, opt);
 161
 162         /*
 163          * Duplicate arguments to shared client group
 164          */
 165         flist_for_each(entry, &arg_list) {
 166                 client = flist_entry(entry, struct fio_client, arg_list);
 167
 168                 __fio_client_add_cmd_option(client, opt);
 169         }
 170 }
 171
 172 struct fio_client *fio_client_add_explicit(const char *hostname, int type,
 173                                            int port)
 174 {
 175         struct fio_client *client;
 176
 177         client = malloc(sizeof(*client));
 178         memset(client, 0, sizeof(*client));
 179
 180         INIT_FLIST_HEAD(&client->list);
 181         INIT_FLIST_HEAD(&client->hash_list);
 182         INIT_FLIST_HEAD(&client->arg_list);
 183         INIT_FLIST_HEAD(&client->eta_list);
 184         INIT_FLIST_HEAD(&client->cmd_list);
 185
 186         client->hostname = strdup(hostname);
 187
 188         if (type == Fio_client_socket)
 189                 client->is_sock = 1;
 190         else {
 191                 int ipv6;
 192
 193                 ipv6 = type == Fio_client_ipv6;
 194                 if (fio_server_parse_host(hostname, &ipv6,
 195                                                 &client->addr.sin_addr,
 196                                                 &client->addr6.sin6_addr))
 197                         goto err;
 198
 199                 client->port = port;
 200         }
 201
 202         client->fd = -1;
 203
 204         __fio_client_add_cmd_option(client, "fio");
 205
 206         flist_add(&client->list, &client_list);
 207         nr_clients++;
 208         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 209         return client;
 210 err:
 211         free(client);
 212         return NULL;
 213 }
 214
 215 int fio_client_add(const char *hostname, void **cookie)
 216 {
 217         struct fio_client *existing = *cookie;
 218         struct fio_client *client;
 219
 220         if (existing) {
 221                 /*
 222                  * We always add our "exec" name as the option, hence 1
 223                  * means empty.
 224                  */
 225                 if (existing->argc == 1)
 226                         flist_add_tail(&existing->arg_list, &arg_list);
 227                 else {
 228                         while (!flist_empty(&arg_list))
 229                                 flist_del_init(arg_list.next);
 230                 }
 231         }
 232
 233         client = malloc(sizeof(*client));
 234         memset(client, 0, sizeof(*client));
 235
 236         INIT_FLIST_HEAD(&client->list);
 237         INIT_FLIST_HEAD(&client->hash_list);
 238         INIT_FLIST_HEAD(&client->arg_list);
 239         INIT_FLIST_HEAD(&client->eta_list);
 240         INIT_FLIST_HEAD(&client->cmd_list);
 241
 242         if (fio_server_parse_string(hostname, &client->hostname,
 243                                         &client->is_sock, &client->port,
 244                                         &client->addr.sin_addr,
 245                                         &client->addr6.sin6_addr,
 246                                         &client->ipv6))
 247                 return -1;
 248
 249         client->fd = -1;
 250
 251         __fio_client_add_cmd_option(client, "fio");
 252
 253         flist_add(&client->list, &client_list);
 254         nr_clients++;
 255         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 256         *cookie = client;
 257         return 0;
 258 }
 259
 260 static int fio_client_connect_ip(struct fio_client *client)
 261 {
 262         struct sockaddr *addr;
 263         fio_socklen_t socklen;
 264         int fd, domain;
 265
 266         if (client->ipv6) {
 267                 client->addr6.sin6_family = AF_INET6;
 268                 client->addr6.sin6_port = htons(client->port);
 269                 domain = AF_INET6;
 270                 addr = (struct sockaddr *) &client->addr6;
 271                 socklen = sizeof(client->addr6);
 272         } else {
 273                 client->addr.sin_family = AF_INET;
 274                 client->addr.sin_port = htons(client->port);
 275                 domain = AF_INET;
 276                 addr = (struct sockaddr *) &client->addr;
 277                 socklen = sizeof(client->addr);
 278         }
 279
 280         fd = socket(domain, SOCK_STREAM, 0);
 281         if (fd < 0) {
 282                 log_err("fio: socket: %s\n", strerror(errno));
 283                 return -1;
 284         }
 285
 286         if (connect(fd, addr, socklen) < 0) {
 287                 log_err("fio: connect: %s\n", strerror(errno));
 288                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 289                                                                 client->port);
 290                 close(fd);
 291                 return -1;
 292         }
 293
 294         return fd;
 295 }
 296
 297 static int fio_client_connect_sock(struct fio_client *client)
 298 {
 299         struct sockaddr_un *addr = &client->addr_un;
 300         fio_socklen_t len;
 301         int fd;
 302
 303         memset(addr, 0, sizeof(*addr));
 304         addr->sun_family = AF_UNIX;
 305         strcpy(addr->sun_path, client->hostname);
 306
 307         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 308         if (fd < 0) {
 309                 log_err("fio: socket: %s\n", strerror(errno));
 310                 return -1;
 311         }
 312
 313         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 314         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 315                 log_err("fio: connect; %s\n", strerror(errno));
 316                 close(fd);
 317                 return -1;
 318         }
 319
 320         return fd;
 321 }
 322
 323 static int fio_client_connect(struct fio_client *client)
 324 {
 325         int fd;
 326
 327         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 328
 329         if (client->is_sock)
 330                 fd = fio_client_connect_sock(client);
 331         else
 332                 fd = fio_client_connect_ip(client);
 333
 334         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 335
 336         if (fd < 0)
 337                 return 1;
 338
 339         client->fd = fd;
 340         fio_client_add_hash(client);
 341         client->state = Client_connected;
 342         return 0;
 343 }
 344
 345 void fio_clients_terminate(void)
 346 {
 347         struct flist_head *entry;
 348         struct fio_client *client;
 349
 350         dprint(FD_NET, "client: terminate clients\n");
 351
 352         flist_for_each(entry, &client_list) {
 353                 client = flist_entry(entry, struct fio_client, list);
 354
 355                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0, NULL);
 356         }
 357 }
 358
 359 static void sig_int(int sig)
 360 {
 361         dprint(FD_NET, "client: got signal %d\n", sig);
 362         fio_clients_terminate();
 363 }
 364
 365 static void client_signal_handler(void)
 366 {
 367         struct sigaction act;
 368
 369         memset(&act, 0, sizeof(act));
 370         act.sa_handler = sig_int;
 371         act.sa_flags = SA_RESTART;
 372         sigaction(SIGINT, &act, NULL);
 373
 374         memset(&act, 0, sizeof(act));
 375         act.sa_handler = sig_int;
 376         act.sa_flags = SA_RESTART;
 377         sigaction(SIGTERM, &act, NULL);
 378 }
 379
 380 static void probe_client(struct fio_client *client)
 381 {
 382         dprint(FD_NET, "client: send probe\n");
 383
 384         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
 385 }
 386
 387 static int send_client_cmd_line(struct fio_client *client)
 388 {
 389         struct cmd_single_line_pdu *cslp;
 390         struct cmd_line_pdu *clp;
 391         unsigned long offset;
 392         unsigned int *lens;
 393         void *pdu;
 394         size_t mem;
 395         int i, ret;
 396
 397         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 398
 399         lens = malloc(client->argc * sizeof(unsigned int));
 400
 401         /*
 402          * Find out how much mem we need
 403          */
 404         for (i = 0, mem = 0; i < client->argc; i++) {
 405                 lens[i] = strlen(client->argv[i]) + 1;
 406                 mem += lens[i];
 407         }
 408
 409         /*
 410          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 411          */
 412         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 413
 414         pdu = malloc(mem);
 415         clp = pdu;
 416         offset = sizeof(*clp);
 417
 418         for (i = 0; i < client->argc; i++) {
 419                 uint16_t arg_len = lens[i];
 420
 421                 cslp = pdu + offset;
 422                 strcpy((char *) cslp->text, client->argv[i]);
 423                 cslp->len = cpu_to_le16(arg_len);
 424                 offset += sizeof(*cslp) + arg_len;
 425         }
 426
 427         free(lens);
 428         clp->lines = cpu_to_le16(client->argc);
 429         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
 430         free(pdu);
 431         return ret;
 432 }
 433
 434 int fio_clients_connect(void)
 435 {
 436         struct fio_client *client;
 437         struct flist_head *entry, *tmp;
 438         int ret;
 439
 440 #ifdef WIN32
 441         WSADATA wsd;
 442         WSAStartup(MAKEWORD(2,2), &wsd);
 443 #endif
 444
 445         dprint(FD_NET, "client: connect all\n");
 446
 447         client_signal_handler();
 448
 449         flist_for_each_safe(entry, tmp, &client_list) {
 450                 client = flist_entry(entry, struct fio_client, list);
 451
 452                 ret = fio_client_connect(client);
 453                 if (ret) {
 454                         remove_client(client);
 455                         continue;
 456                 }
 457
 458                 probe_client(client);
 459
 460                 if (client->argc > 1)
 461                         send_client_cmd_line(client);
 462         }
 463
 464         return !nr_clients;
 465 }
 466
 467 /*
 468  * Send file contents to server backend. We could use sendfile(), but to remain
 469  * more portable lets just read/write the darn thing.
 470  */
 471 static int fio_client_send_ini(struct fio_client *client, const char *filename)
 472 {
 473         struct stat sb;
 474         char *p, *buf;
 475         off_t len;
 476         int fd, ret;
 477
 478         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 479
 480         fd = open(filename, O_RDONLY);
 481         if (fd < 0) {
 482                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 483                 return 1;
 484         }
 485
 486         if (fstat(fd, &sb) < 0) {
 487                 log_err("fio: job file stat: %s\n", strerror(errno));
 488                 close(fd);
 489                 return 1;
 490         }
 491
 492         buf = malloc(sb.st_size);
 493
 494         len = sb.st_size;
 495         p = buf;
 496         do {
 497                 ret = read(fd, p, len);
 498                 if (ret > 0) {
 499                         len -= ret;
 500                         if (!len)
 501                                 break;
 502                         p += ret;
 503                         continue;
 504                 } else if (!ret)
 505                         break;
 506                 else if (errno == EAGAIN || errno == EINTR)
 507                         continue;
 508         } while (1);
 509
 510         if (len) {
 511                 log_err("fio: failed reading job file %s\n", filename);
 512                 close(fd);
 513                 free(buf);
 514                 return 1;
 515         }
 516
 517         client->sent_job = 1;
 518         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size, 0);
 519         free(buf);
 520         close(fd);
 521         return ret;
 522 }
 523
 524 int fio_clients_send_ini(const char *filename)
 525 {
 526         struct fio_client *client;
 527         struct flist_head *entry, *tmp;
 528
 529         flist_for_each_safe(entry, tmp, &client_list) {
 530                 client = flist_entry(entry, struct fio_client, list);
 531
 532                 if (fio_client_send_ini(client, filename))
 533                         remove_client(client);
 534
 535                 client->sent_job = 1;
 536         }
 537
 538         return !nr_clients;
 539 }
 540
 541 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 542 {
 543         dst->max_val    = le64_to_cpu(src->max_val);
 544         dst->min_val    = le64_to_cpu(src->min_val);
 545         dst->samples    = le64_to_cpu(src->samples);
 546
 547         /*
 548          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 549          */
 550         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 551         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 552 }
 553
 554 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 555 {
 556         int i, j;
 557
 558         dst->error      = le32_to_cpu(src->error);
 559         dst->groupid    = le32_to_cpu(src->groupid);
 560         dst->pid        = le32_to_cpu(src->pid);
 561         dst->members    = le32_to_cpu(src->members);
 562
 563         for (i = 0; i < 2; i++) {
 564                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 565                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 566                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 567                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 568         }
 569
 570         dst->usr_time           = le64_to_cpu(src->usr_time);
 571         dst->sys_time           = le64_to_cpu(src->sys_time);
 572         dst->ctx                = le64_to_cpu(src->ctx);
 573         dst->minf               = le64_to_cpu(src->minf);
 574         dst->majf               = le64_to_cpu(src->majf);
 575         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 576
 577         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 578                 fio_fp64_t *fps = &src->percentile_list[i];
 579                 fio_fp64_t *fpd = &dst->percentile_list[i];
 580
 581                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 582         }
 583
 584         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 585                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 586                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 587                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 588         }
 589
 590         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 591                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 592                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 593         }
 594
 595         for (i = 0; i < 2; i++)
 596                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 597                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 598
 599         for (i = 0; i < 3; i++) {
 600                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 601                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 602         }
 603
 604         dst->total_submit       = le64_to_cpu(src->total_submit);
 605         dst->total_complete     = le64_to_cpu(src->total_complete);
 606
 607         for (i = 0; i < 2; i++) {
 608                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 609                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 610         }
 611
 612         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 613         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 614         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 615         dst->first_error        = le32_to_cpu(src->first_error);
 616         dst->kb_base            = le32_to_cpu(src->kb_base);
 617 }
 618
 619 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 620 {
 621         int i;
 622
 623         for (i = 0; i < 2; i++) {
 624                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 625                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 626                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 627                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 628                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 629                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 630         }
 631
 632         dst->kb_base    = le32_to_cpu(src->kb_base);
 633         dst->groupid    = le32_to_cpu(src->groupid);
 634 }
 635
 636 static void handle_ts(struct fio_net_cmd *cmd)
 637 {
 638         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 639
 640         convert_ts(&p->ts, &p->ts);
 641         convert_gs(&p->rs, &p->rs);
 642
 643         show_thread_status(&p->ts, &p->rs);
 644
 645         if (sum_stat_clients == 1)
 646                 return;
 647
 648         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 649         sum_group_stats(&client_gs, &p->rs);
 650
 651         client_ts.members++;
 652         client_ts.groupid = p->ts.groupid;
 653
 654         if (++sum_stat_nr == sum_stat_clients) {
 655                 strcpy(client_ts.name, "All clients");
 656                 show_thread_status(&client_ts, &client_gs);
 657         }
 658 }
 659
 660 static void handle_gs(struct fio_net_cmd *cmd)
 661 {
 662         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 663
 664         convert_gs(gs, gs);
 665         show_group_stats(gs);
 666 }
 667
 668 static void convert_agg(struct disk_util_agg *agg)
 669 {
 670         int i;
 671
 672         for (i = 0; i < 2; i++) {
 673                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 674                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 675                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 676                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 677         }
 678
 679         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 680         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 681         agg->slavecount         = le32_to_cpu(agg->slavecount);
 682         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 683 }
 684
 685 static void convert_dus(struct disk_util_stat *dus)
 686 {
 687         int i;
 688
 689         for (i = 0; i < 2; i++) {
 690                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
 691                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
 692                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
 693                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
 694         }
 695
 696         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
 697         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
 698         dus->msec               = le64_to_cpu(dus->msec);
 699 }
 700
 701 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 702 {
 703         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 704
 705         convert_dus(&du->dus);
 706         convert_agg(&du->agg);
 707
 708         if (!client->disk_stats_shown) {
 709                 client->disk_stats_shown = 1;
 710                 log_info("\nDisk stats (read/write):\n");
 711         }
 712
 713         print_disk_util(&du->dus, &du->agg, terse_output);
 714 }
 715
 716 void fio_client_convert_jobs_eta(struct jobs_eta *je)
 717 {
 718         int i;
 719
 720         je->nr_running          = le32_to_cpu(je->nr_running);
 721         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 722         je->nr_pending          = le32_to_cpu(je->nr_pending);
 723         je->files_open          = le32_to_cpu(je->files_open);
 724
 725         for (i = 0; i < 2; i++) {
 726                 je->m_rate[i]           = le32_to_cpu(je->m_rate[i]);
 727                 je->t_rate[i]           = le32_to_cpu(je->t_rate[i]);
 728                 je->m_iops[i]           = le32_to_cpu(je->m_iops[i]);
 729                 je->t_iops[i]           = le32_to_cpu(je->t_iops[i]);
 730                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 731                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 732         }
 733
 734         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 735         je->eta_sec             = le64_to_cpu(je->eta_sec);
 736 }
 737
 738 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 739 {
 740         int i;
 741
 742         dst->nr_running         += je->nr_running;
 743         dst->nr_ramp            += je->nr_ramp;
 744         dst->nr_pending         += je->nr_pending;
 745         dst->files_open         += je->files_open;
 746
 747         for (i = 0; i < 2; i++) {
 748                 dst->m_rate[i]  += je->m_rate[i];
 749                 dst->t_rate[i]  += je->t_rate[i];
 750                 dst->m_iops[i]  += je->m_iops[i];
 751                 dst->t_iops[i]  += je->t_iops[i];
 752                 dst->rate[i]    += je->rate[i];
 753                 dst->iops[i]    += je->iops[i];
 754         }
 755
 756         dst->elapsed_sec        += je->elapsed_sec;
 757
 758         if (je->eta_sec > dst->eta_sec)
 759                 dst->eta_sec = je->eta_sec;
 760 }
 761
 762 void fio_client_dec_jobs_eta(struct client_eta *eta, void (*fn)(struct jobs_eta *))
 763 {
 764         if (!--eta->pending) {
 765                 fn(&eta->eta);
 766                 free(eta);
 767         }
 768 }
 769
 770 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 771 {
 772         struct fio_net_int_cmd *icmd = NULL;
 773         struct flist_head *entry;
 774
 775         flist_for_each(entry, &client->cmd_list) {
 776                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
 777
 778                 if (cmd->tag == (uintptr_t) icmd)
 779                         break;
 780
 781                 icmd = NULL;
 782         }
 783
 784         if (!icmd) {
 785                 log_err("fio: client: unable to find matching tag\n");
 786                 return;
 787         }
 788
 789         flist_del(&icmd->list);
 790         cmd->tag = icmd->saved_tag;
 791         free(icmd);
 792 }
 793
 794 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 795 {
 796         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 797         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 798
 799         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 800
 801         assert(client->eta_in_flight == eta);
 802
 803         client->eta_in_flight = NULL;
 804         flist_del_init(&client->eta_list);
 805
 806         fio_client_convert_jobs_eta(je);
 807         fio_client_sum_jobs_eta(&eta->eta, je);
 808         fio_client_dec_jobs_eta(eta, display_thread_status);
 809 }
 810
 811 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 812 {
 813         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 814         const char *os, *arch;
 815         char bit[16];
 816
 817         os = fio_get_os_string(probe->os);
 818         if (!os)
 819                 os = "unknown";
 820
 821         arch = fio_get_arch_string(probe->arch);
 822         if (!arch)
 823                 os = "unknown";
 824
 825         sprintf(bit, "%d-bit", probe->bpp * 8);
 826
 827         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
 828                 probe->hostname, probe->bigendian, bit, os, arch,
 829                 probe->fio_major, probe->fio_minor, probe->fio_patch);
 830
 831         if (!client->name)
 832                 client->name = strdup((char *) probe->hostname);
 833 }
 834
 835 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
 836 {
 837         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 838
 839         client->state = Client_started;
 840         client->jobs = le32_to_cpu(pdu->jobs);
 841 }
 842
 843 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
 844 {
 845         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
 846
 847         client->state = Client_stopped;
 848         client->error = le32_to_cpu(pdu->error);
 849
 850         if (client->error)
 851                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
 852 }
 853
 854 int fio_handle_client(struct fio_client *client, struct client_ops *ops)
 855 {
 856         struct fio_net_cmd *cmd;
 857
 858         dprint(FD_NET, "client: handle %s\n", client->hostname);
 859
 860         cmd = fio_net_recv_cmd(client->fd);
 861         if (!cmd)
 862                 return 0;
 863
 864         dprint(FD_NET, "client: got cmd op %s from %s\n",
 865                                 fio_server_op(cmd->opcode), client->hostname);
 866
 867         switch (cmd->opcode) {
 868         case FIO_NET_CMD_QUIT:
 869                 if (ops->quit)
 870                         ops->quit(client);
 871                 remove_client(client);
 872                 free(cmd);
 873                 break;
 874         case FIO_NET_CMD_TEXT: {
 875                 const char *buf = (const char *) cmd->payload;
 876                 ops->text_op(client, f_out, cmd->pdu_len, buf);
 877                 free(cmd);
 878                 break;
 879                 }
 880         case FIO_NET_CMD_DU:
 881                 ops->disk_util(client, cmd);
 882                 free(cmd);
 883                 break;
 884         case FIO_NET_CMD_TS:
 885                 ops->thread_status(cmd);
 886                 free(cmd);
 887                 break;
 888         case FIO_NET_CMD_GS:
 889                 ops->group_stats(cmd);
 890                 free(cmd);
 891                 break;
 892         case FIO_NET_CMD_ETA:
 893                 remove_reply_cmd(client, cmd);
 894                 ops->eta(client, cmd);
 895                 free(cmd);
 896                 break;
 897         case FIO_NET_CMD_PROBE:
 898                 remove_reply_cmd(client, cmd);
 899                 ops->probe(client, cmd);
 900                 free(cmd);
 901                 break;
 902         case FIO_NET_CMD_RUN:
 903                 client->state = Client_running;
 904                 free(cmd);
 905                 break;
 906         case FIO_NET_CMD_START:
 907                 handle_start(client, cmd);
 908                 free(cmd);
 909                 break;
 910         case FIO_NET_CMD_STOP:
 911                 handle_stop(client, cmd);
 912                 free(cmd);
 913                 break;
 914         case FIO_NET_CMD_ADD_JOB:
 915                 if (ops->add_job)
 916                         ops->add_job(client, cmd);
 917                 free(cmd);
 918                 break;
 919         default:
 920                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
 921                 free(cmd);
 922                 break;
 923         }
 924
 925         return 1;
 926 }
 927
 928 static void request_client_etas(void)
 929 {
 930         struct fio_client *client;
 931         struct flist_head *entry;
 932         struct client_eta *eta;
 933         int skipped = 0;
 934
 935         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
 936
 937         eta = malloc(sizeof(*eta));
 938         memset(&eta->eta, 0, sizeof(eta->eta));
 939         eta->pending = nr_clients;
 940
 941         flist_for_each(entry, &client_list) {
 942                 client = flist_entry(entry, struct fio_client, list);
 943
 944                 if (!flist_empty(&client->eta_list)) {
 945                         skipped++;
 946                         continue;
 947                 }
 948                 if (client->state != Client_running)
 949                         continue;
 950
 951                 assert(!client->eta_in_flight);
 952                 flist_add_tail(&client->eta_list, &eta_list);
 953                 client->eta_in_flight = eta;
 954                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
 955                                         (uintptr_t) eta, &client->cmd_list);
 956         }
 957
 958         while (skipped--)
 959                 fio_client_dec_jobs_eta(eta, display_thread_status);
 960
 961         dprint(FD_NET, "client: requested eta tag %p\n", eta);
 962 }
 963
 964 static int client_check_cmd_timeout(struct fio_client *client,
 965                                     struct timeval *now)
 966 {
 967         struct fio_net_int_cmd *cmd;
 968         struct flist_head *entry, *tmp;
 969         int ret = 0;
 970
 971         flist_for_each_safe(entry, tmp, &client->cmd_list) {
 972                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
 973
 974                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
 975                         continue;
 976
 977                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
 978                                                 fio_server_op(cmd->cmd.opcode));
 979                 flist_del(&cmd->list);
 980                 free(cmd);
 981                 ret = 1;
 982         }
 983
 984         return flist_empty(&client->cmd_list) && ret;
 985 }
 986
 987 static int fio_client_timed_out(struct client_ops *ops)
 988 {
 989         struct fio_client *client;
 990         struct flist_head *entry, *tmp;
 991         struct timeval tv;
 992         int ret = 0;
 993
 994         gettimeofday(&tv, NULL);
 995
 996         flist_for_each_safe(entry, tmp, &client_list) {
 997                 client = flist_entry(entry, struct fio_client, list);
 998
 999                 if (flist_empty(&client->cmd_list))
1000                         continue;
1001
1002                 if (!client_check_cmd_timeout(client, &tv))
1003                         continue;
1004
1005                 if (ops->timed_out)
1006                         ops->timed_out(client);
1007                 else
1008                         log_err("fio: client %s timed out\n", client->hostname);
1009
1010                 remove_client(client);
1011                 ret = 1;
1012         }
1013
1014         return ret;
1015 }
1016
1017 int fio_handle_clients(struct client_ops *ops)
1018 {
1019         struct pollfd *pfds;
1020         int i, ret = 0, retval = 0;
1021
1022         gettimeofday(&eta_tv, NULL);
1023
1024         pfds = malloc(nr_clients * sizeof(struct pollfd));
1025
1026         sum_stat_clients = nr_clients;
1027         init_thread_stat(&client_ts);
1028         init_group_run_stat(&client_gs);
1029
1030         while (!exit_backend && nr_clients) {
1031                 struct flist_head *entry, *tmp;
1032                 struct fio_client *client;
1033
1034                 i = 0;
1035                 flist_for_each_safe(entry, tmp, &client_list) {
1036                         client = flist_entry(entry, struct fio_client, list);
1037
1038                         if (!client->sent_job && !ops->stay_connected &&
1039                             flist_empty(&client->cmd_list)) {
1040                                 remove_client(client);
1041                                 continue;
1042                         }
1043
1044                         pfds[i].fd = client->fd;
1045                         pfds[i].events = POLLIN;
1046                         i++;
1047                 }
1048
1049                 if (!nr_clients)
1050                         break;
1051
1052                 assert(i == nr_clients);
1053
1054                 do {
1055                         struct timeval tv;
1056
1057                         gettimeofday(&tv, NULL);
1058                         if (mtime_since(&eta_tv, &tv) >= 900) {
1059                                 request_client_etas();
1060                                 memcpy(&eta_tv, &tv, sizeof(tv));
1061
1062                                 if (fio_client_timed_out(ops))
1063                                         break;
1064                         }
1065
1066                         ret = poll(pfds, nr_clients, 100);
1067                         if (ret < 0) {
1068                                 if (errno == EINTR)
1069                                         continue;
1070                                 log_err("fio: poll clients: %s\n", strerror(errno));
1071                                 break;
1072                         } else if (!ret)
1073                                 continue;
1074                 } while (ret <= 0);
1075
1076                 for (i = 0; i < nr_clients; i++) {
1077                         if (!(pfds[i].revents & POLLIN))
1078                                 continue;
1079
1080                         client = find_client_by_fd(pfds[i].fd);
1081                         if (!client) {
1082                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1083                                 continue;
1084                         }
1085                         if (!fio_handle_client(client, ops)) {
1086                                 log_info("client: host=%s disconnected\n",
1087                                                 client->hostname);
1088                                 remove_client(client);
1089                                 retval = 1;
1090                         } else if (client->error)
1091                                 retval = 1;
1092                 }
1093         }
1094
1095         free(pfds);
1096         return retval;
1097 }