client.c

   1 #include <stdio.h>
   2 #include <stdlib.h>
   3 #include <unistd.h>
   4 #include <limits.h>
   5 #include <errno.h>
   6 #include <fcntl.h>
   7 #include <sys/poll.h>
   8 #include <sys/types.h>
   9 #include <sys/stat.h>
  10 #include <sys/wait.h>
  11 #include <sys/socket.h>
  12 #include <sys/un.h>
  13 #include <netinet/in.h>
  14 #include <arpa/inet.h>
  15 #include <netdb.h>
  16 #include <signal.h>
  17
  18 #include "fio.h"
  19 #include "server.h"
  20 #include "flist.h"
  21 #include "hash.h"
  22
  23 struct client_eta {
  24         struct jobs_eta eta;
  25         unsigned int pending;
  26 };
  27
  28 struct fio_client {
  29         struct flist_head list;
  30         struct flist_head hash_list;
  31         struct flist_head arg_list;
  32         union {
  33                 struct sockaddr_in addr;
  34                 struct sockaddr_in6 addr6;
  35                 struct sockaddr_un addr_un;
  36         };
  37         char *hostname;
  38         int port;
  39         int fd;
  40
  41         char *name;
  42
  43         int state;
  44
  45         int skip_newline;
  46         int is_sock;
  47         int disk_stats_shown;
  48         unsigned int jobs;
  49         int error;
  50         int ipv6;
  51
  52         struct flist_head eta_list;
  53         struct client_eta *eta_in_flight;
  54
  55         struct flist_head cmd_list;
  56
  57         uint16_t argc;
  58         char **argv;
  59 };
  60
  61 static struct timeval eta_tv;
  62
  63 enum {
  64         Client_created          = 0,
  65         Client_connected        = 1,
  66         Client_started          = 2,
  67         Client_running          = 3,
  68         Client_stopped          = 4,
  69         Client_exited           = 5,
  70 };
  71
  72 static FLIST_HEAD(client_list);
  73 static FLIST_HEAD(eta_list);
  74
  75 static FLIST_HEAD(arg_list);
  76
  77 static struct thread_stat client_ts;
  78 static struct group_run_stats client_gs;
  79 static int sum_stat_clients;
  80 static int sum_stat_nr;
  81
  82 #define FIO_CLIENT_HASH_BITS    7
  83 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  84 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  85 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  86
  87 static int handle_client(struct fio_client *client);
  88 static void dec_jobs_eta(struct client_eta *eta);
  89
  90 static void fio_client_add_hash(struct fio_client *client)
  91 {
  92         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  93
  94         bucket &= FIO_CLIENT_HASH_MASK;
  95         flist_add(&client->hash_list, &client_hash[bucket]);
  96 }
  97
  98 static void fio_client_remove_hash(struct fio_client *client)
  99 {
 100         if (!flist_empty(&client->hash_list))
 101                 flist_del_init(&client->hash_list);
 102 }
 103
 104 static void fio_init fio_client_hash_init(void)
 105 {
 106         int i;
 107
 108         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
 109                 INIT_FLIST_HEAD(&client_hash[i]);
 110 }
 111
 112 static struct fio_client *find_client_by_fd(int fd)
 113 {
 114         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
 115         struct fio_client *client;
 116         struct flist_head *entry;
 117
 118         flist_for_each(entry, &client_hash[bucket]) {
 119                 client = flist_entry(entry, struct fio_client, hash_list);
 120
 121                 if (client->fd == fd)
 122                         return client;
 123         }
 124
 125         return NULL;
 126 }
 127
 128 static void remove_client(struct fio_client *client)
 129 {
 130         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 131         flist_del(&client->list);
 132
 133         fio_client_remove_hash(client);
 134
 135         if (!flist_empty(&client->eta_list)) {
 136                 flist_del_init(&client->eta_list);
 137                 dec_jobs_eta(client->eta_in_flight);
 138         }
 139
 140         free(client->hostname);
 141         if (client->argv)
 142                 free(client->argv);
 143         if (client->name)
 144                 free(client->name);
 145
 146         free(client);
 147         nr_clients--;
 148         sum_stat_clients--;
 149 }
 150
 151 static void __fio_client_add_cmd_option(struct fio_client *client,
 152                                         const char *opt)
 153 {
 154         int index;
 155
 156         index = client->argc++;
 157         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 158         client->argv[index] = strdup(opt);
 159         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 160 }
 161
 162 void fio_client_add_cmd_option(void *cookie, const char *opt)
 163 {
 164         struct fio_client *client = cookie;
 165         struct flist_head *entry;
 166
 167         if (!client || !opt)
 168                 return;
 169
 170         __fio_client_add_cmd_option(client, opt);
 171
 172         /*
 173          * Duplicate arguments to shared client group
 174          */
 175         flist_for_each(entry, &arg_list) {
 176                 client = flist_entry(entry, struct fio_client, arg_list);
 177
 178                 __fio_client_add_cmd_option(client, opt);
 179         }
 180 }
 181
 182 int fio_client_add(const char *hostname, void **cookie)
 183 {
 184         struct fio_client *existing = *cookie;
 185         struct fio_client *client;
 186
 187         if (existing) {
 188                 /*
 189                  * We always add our "exec" name as the option, hence 1
 190                  * means empty.
 191                  */
 192                 if (existing->argc == 1)
 193                         flist_add_tail(&existing->arg_list, &arg_list);
 194                 else {
 195                         while (!flist_empty(&arg_list))
 196                                 flist_del_init(arg_list.next);
 197                 }
 198         }
 199
 200         client = malloc(sizeof(*client));
 201         memset(client, 0, sizeof(*client));
 202
 203         INIT_FLIST_HEAD(&client->list);
 204         INIT_FLIST_HEAD(&client->hash_list);
 205         INIT_FLIST_HEAD(&client->arg_list);
 206         INIT_FLIST_HEAD(&client->eta_list);
 207         INIT_FLIST_HEAD(&client->cmd_list);
 208
 209         if (fio_server_parse_string(hostname, &client->hostname,
 210                                         &client->is_sock, &client->port,
 211                                         &client->addr.sin_addr,
 212                                         &client->addr6.sin6_addr,
 213                                         &client->ipv6))
 214                 return -1;
 215
 216         client->fd = -1;
 217
 218         __fio_client_add_cmd_option(client, "fio");
 219
 220         flist_add(&client->list, &client_list);
 221         nr_clients++;
 222         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 223         *cookie = client;
 224         return 0;
 225 }
 226
 227 static int fio_client_connect_ip(struct fio_client *client)
 228 {
 229         struct sockaddr *addr;
 230         fio_socklen_t socklen;
 231         int fd, domain;
 232
 233         if (client->ipv6) {
 234                 client->addr6.sin6_family = AF_INET6;
 235                 client->addr6.sin6_port = htons(client->port);
 236                 domain = AF_INET6;
 237                 addr = (struct sockaddr *) &client->addr6;
 238                 socklen = sizeof(client->addr6);
 239         } else {
 240                 client->addr.sin_family = AF_INET;
 241                 client->addr.sin_port = htons(client->port);
 242                 domain = AF_INET;
 243                 addr = (struct sockaddr *) &client->addr;
 244                 socklen = sizeof(client->addr);
 245         }
 246
 247         fd = socket(domain, SOCK_STREAM, 0);
 248         if (fd < 0) {
 249                 log_err("fio: socket: %s\n", strerror(errno));
 250                 return -1;
 251         }
 252
 253         if (connect(fd, addr, socklen) < 0) {
 254                 log_err("fio: connect: %s\n", strerror(errno));
 255                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 256                                                                 client->port);
 257                 close(fd);
 258                 return -1;
 259         }
 260
 261         return fd;
 262 }
 263
 264 static int fio_client_connect_sock(struct fio_client *client)
 265 {
 266         struct sockaddr_un *addr = &client->addr_un;
 267         fio_socklen_t len;
 268         int fd;
 269
 270         memset(addr, 0, sizeof(*addr));
 271         addr->sun_family = AF_UNIX;
 272         strcpy(addr->sun_path, client->hostname);
 273
 274         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 275         if (fd < 0) {
 276                 log_err("fio: socket: %s\n", strerror(errno));
 277                 return -1;
 278         }
 279
 280         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 281         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 282                 log_err("fio: connect; %s\n", strerror(errno));
 283                 close(fd);
 284                 return -1;
 285         }
 286
 287         return fd;
 288 }
 289
 290 static int fio_client_connect(struct fio_client *client)
 291 {
 292         int fd;
 293
 294         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 295
 296         if (client->is_sock)
 297                 fd = fio_client_connect_sock(client);
 298         else
 299                 fd = fio_client_connect_ip(client);
 300
 301         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 302
 303         if (fd < 0)
 304                 return 1;
 305
 306         client->fd = fd;
 307         fio_client_add_hash(client);
 308         client->state = Client_connected;
 309         return 0;
 310 }
 311
 312 void fio_clients_terminate(void)
 313 {
 314         struct flist_head *entry;
 315         struct fio_client *client;
 316
 317         dprint(FD_NET, "client: terminate clients\n");
 318
 319         flist_for_each(entry, &client_list) {
 320                 client = flist_entry(entry, struct fio_client, list);
 321
 322                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0, NULL);
 323         }
 324 }
 325
 326 static void sig_int(int sig)
 327 {
 328         dprint(FD_NET, "client: got signal %d\n", sig);
 329         fio_clients_terminate();
 330 }
 331
 332 static void client_signal_handler(void)
 333 {
 334         struct sigaction act;
 335
 336         memset(&act, 0, sizeof(act));
 337         act.sa_handler = sig_int;
 338         act.sa_flags = SA_RESTART;
 339         sigaction(SIGINT, &act, NULL);
 340
 341         memset(&act, 0, sizeof(act));
 342         act.sa_handler = sig_int;
 343         act.sa_flags = SA_RESTART;
 344         sigaction(SIGTERM, &act, NULL);
 345 }
 346
 347 static void probe_client(struct fio_client *client)
 348 {
 349         dprint(FD_NET, "client: send probe\n");
 350
 351         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
 352 }
 353
 354 static int send_client_cmd_line(struct fio_client *client)
 355 {
 356         struct cmd_single_line_pdu *cslp;
 357         struct cmd_line_pdu *clp;
 358         unsigned long offset;
 359         unsigned int *lens;
 360         void *pdu;
 361         size_t mem;
 362         int i, ret;
 363
 364         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 365
 366         lens = malloc(client->argc * sizeof(unsigned int));
 367
 368         /*
 369          * Find out how much mem we need
 370          */
 371         for (i = 0, mem = 0; i < client->argc; i++) {
 372                 lens[i] = strlen(client->argv[i]) + 1;
 373                 mem += lens[i];
 374         }
 375
 376         /*
 377          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 378          */
 379         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 380
 381         pdu = malloc(mem);
 382         clp = pdu;
 383         offset = sizeof(*clp);
 384
 385         for (i = 0; i < client->argc; i++) {
 386                 uint16_t arg_len = lens[i];
 387
 388                 cslp = pdu + offset;
 389                 strcpy((char *) cslp->text, client->argv[i]);
 390                 cslp->len = cpu_to_le16(arg_len);
 391                 offset += sizeof(*cslp) + arg_len;
 392         }
 393
 394         free(lens);
 395         clp->lines = cpu_to_le16(client->argc);
 396         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
 397         free(pdu);
 398         return ret;
 399 }
 400
 401 int fio_clients_connect(void)
 402 {
 403         struct fio_client *client;
 404         struct flist_head *entry, *tmp;
 405         int ret;
 406
 407 #ifdef WIN32
 408         WSADATA wsd;
 409         WSAStartup(MAKEWORD(2,2), &wsd);
 410 #endif
 411
 412         dprint(FD_NET, "client: connect all\n");
 413
 414         client_signal_handler();
 415
 416         flist_for_each_safe(entry, tmp, &client_list) {
 417                 client = flist_entry(entry, struct fio_client, list);
 418
 419                 ret = fio_client_connect(client);
 420                 if (ret) {
 421                         remove_client(client);
 422                         continue;
 423                 }
 424
 425                 probe_client(client);
 426
 427                 if (client->argc > 1)
 428                         send_client_cmd_line(client);
 429         }
 430
 431         return !nr_clients;
 432 }
 433
 434 /*
 435  * Send file contents to server backend. We could use sendfile(), but to remain
 436  * more portable lets just read/write the darn thing.
 437  */
 438 static int fio_client_send_ini(struct fio_client *client, const char *filename)
 439 {
 440         struct stat sb;
 441         char *p, *buf;
 442         off_t len;
 443         int fd, ret;
 444
 445         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 446
 447         fd = open(filename, O_RDONLY);
 448         if (fd < 0) {
 449                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 450                 return 1;
 451         }
 452
 453         if (fstat(fd, &sb) < 0) {
 454                 log_err("fio: job file stat: %s\n", strerror(errno));
 455                 close(fd);
 456                 return 1;
 457         }
 458
 459         buf = malloc(sb.st_size);
 460
 461         len = sb.st_size;
 462         p = buf;
 463         do {
 464                 ret = read(fd, p, len);
 465                 if (ret > 0) {
 466                         len -= ret;
 467                         if (!len)
 468                                 break;
 469                         p += ret;
 470                         continue;
 471                 } else if (!ret)
 472                         break;
 473                 else if (errno == EAGAIN || errno == EINTR)
 474                         continue;
 475         } while (1);
 476
 477         if (len) {
 478                 log_err("fio: failed reading job file %s\n", filename);
 479                 close(fd);
 480                 free(buf);
 481                 return 1;
 482         }
 483
 484         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size, 0);
 485         free(buf);
 486         close(fd);
 487         return ret;
 488 }
 489
 490 int fio_clients_send_ini(const char *filename)
 491 {
 492         struct fio_client *client;
 493         struct flist_head *entry, *tmp;
 494
 495         flist_for_each_safe(entry, tmp, &client_list) {
 496                 client = flist_entry(entry, struct fio_client, list);
 497
 498                 if (fio_client_send_ini(client, filename))
 499                         remove_client(client);
 500         }
 501
 502         return !nr_clients;
 503 }
 504
 505 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 506 {
 507         dst->max_val    = le64_to_cpu(src->max_val);
 508         dst->min_val    = le64_to_cpu(src->min_val);
 509         dst->samples    = le64_to_cpu(src->samples);
 510
 511         /*
 512          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 513          */
 514         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 515         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 516 }
 517
 518 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 519 {
 520         int i, j;
 521
 522         dst->error      = le32_to_cpu(src->error);
 523         dst->groupid    = le32_to_cpu(src->groupid);
 524         dst->pid        = le32_to_cpu(src->pid);
 525         dst->members    = le32_to_cpu(src->members);
 526
 527         for (i = 0; i < 2; i++) {
 528                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 529                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 530                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 531                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 532         }
 533
 534         dst->usr_time           = le64_to_cpu(src->usr_time);
 535         dst->sys_time           = le64_to_cpu(src->sys_time);
 536         dst->ctx                = le64_to_cpu(src->ctx);
 537         dst->minf               = le64_to_cpu(src->minf);
 538         dst->majf               = le64_to_cpu(src->majf);
 539         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 540
 541         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 542                 fio_fp64_t *fps = &src->percentile_list[i];
 543                 fio_fp64_t *fpd = &dst->percentile_list[i];
 544
 545                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 546         }
 547
 548         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 549                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 550                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 551                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 552         }
 553
 554         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 555                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 556                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 557         }
 558
 559         for (i = 0; i < 2; i++)
 560                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 561                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 562
 563         for (i = 0; i < 3; i++) {
 564                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 565                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 566         }
 567
 568         dst->total_submit       = le64_to_cpu(src->total_submit);
 569         dst->total_complete     = le64_to_cpu(src->total_complete);
 570
 571         for (i = 0; i < 2; i++) {
 572                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 573                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 574         }
 575
 576         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 577         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 578         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 579         dst->first_error        = le32_to_cpu(src->first_error);
 580         dst->kb_base            = le32_to_cpu(src->kb_base);
 581 }
 582
 583 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 584 {
 585         int i;
 586
 587         for (i = 0; i < 2; i++) {
 588                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 589                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 590                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 591                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 592                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 593                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 594         }
 595
 596         dst->kb_base    = le32_to_cpu(src->kb_base);
 597         dst->groupid    = le32_to_cpu(src->groupid);
 598 }
 599
 600 static void handle_ts(struct fio_net_cmd *cmd)
 601 {
 602         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 603
 604         convert_ts(&p->ts, &p->ts);
 605         convert_gs(&p->rs, &p->rs);
 606
 607         show_thread_status(&p->ts, &p->rs);
 608
 609         if (sum_stat_clients == 1)
 610                 return;
 611
 612         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 613         sum_group_stats(&client_gs, &p->rs);
 614
 615         client_ts.members++;
 616         client_ts.groupid = p->ts.groupid;
 617
 618         if (++sum_stat_nr == sum_stat_clients) {
 619                 strcpy(client_ts.name, "All clients");
 620                 show_thread_status(&client_ts, &client_gs);
 621         }
 622 }
 623
 624 static void handle_gs(struct fio_net_cmd *cmd)
 625 {
 626         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 627
 628         convert_gs(gs, gs);
 629         show_group_stats(gs);
 630 }
 631
 632 static void convert_agg(struct disk_util_agg *agg)
 633 {
 634         int i;
 635
 636         for (i = 0; i < 2; i++) {
 637                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 638                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 639                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 640                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 641         }
 642
 643         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 644         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 645         agg->slavecount         = le32_to_cpu(agg->slavecount);
 646         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 647 }
 648
 649 static void convert_dus(struct disk_util_stat *dus)
 650 {
 651         int i;
 652
 653         for (i = 0; i < 2; i++) {
 654                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
 655                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
 656                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
 657                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
 658         }
 659
 660         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
 661         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
 662         dus->msec               = le64_to_cpu(dus->msec);
 663 }
 664
 665 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 666 {
 667         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 668
 669         convert_dus(&du->dus);
 670         convert_agg(&du->agg);
 671
 672         if (!client->disk_stats_shown) {
 673                 client->disk_stats_shown = 1;
 674                 log_info("\nDisk stats (read/write):\n");
 675         }
 676
 677         print_disk_util(&du->dus, &du->agg, terse_output);
 678 }
 679
 680 static void convert_jobs_eta(struct jobs_eta *je)
 681 {
 682         int i;
 683
 684         je->nr_running          = le32_to_cpu(je->nr_running);
 685         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 686         je->nr_pending          = le32_to_cpu(je->nr_pending);
 687         je->files_open          = le32_to_cpu(je->files_open);
 688         je->m_rate              = le32_to_cpu(je->m_rate);
 689         je->t_rate              = le32_to_cpu(je->t_rate);
 690         je->m_iops              = le32_to_cpu(je->m_iops);
 691         je->t_iops              = le32_to_cpu(je->t_iops);
 692
 693         for (i = 0; i < 2; i++) {
 694                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 695                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 696         }
 697
 698         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 699         je->eta_sec             = le64_to_cpu(je->eta_sec);
 700 }
 701
 702 static void sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 703 {
 704         int i;
 705
 706         dst->nr_running         += je->nr_running;
 707         dst->nr_ramp            += je->nr_ramp;
 708         dst->nr_pending         += je->nr_pending;
 709         dst->files_open         += je->files_open;
 710         dst->m_rate             += je->m_rate;
 711         dst->t_rate             += je->t_rate;
 712         dst->m_iops             += je->m_iops;
 713         dst->t_iops             += je->t_iops;
 714
 715         for (i = 0; i < 2; i++) {
 716                 dst->rate[i]    += je->rate[i];
 717                 dst->iops[i]    += je->iops[i];
 718         }
 719
 720         dst->elapsed_sec        += je->elapsed_sec;
 721
 722         if (je->eta_sec > dst->eta_sec)
 723                 dst->eta_sec = je->eta_sec;
 724 }
 725
 726 static void dec_jobs_eta(struct client_eta *eta)
 727 {
 728         if (!--eta->pending) {
 729                 display_thread_status(&eta->eta);
 730                 free(eta);
 731         }
 732 }
 733
 734 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 735 {
 736         struct fio_net_int_cmd *icmd = NULL;
 737         struct flist_head *entry;
 738
 739         flist_for_each(entry, &client->cmd_list) {
 740                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
 741
 742                 if (cmd->tag == (uintptr_t) icmd)
 743                         break;
 744
 745                 icmd = NULL;
 746         }
 747
 748         if (!icmd) {
 749                 log_err("fio: client: unable to find matching tag\n");
 750                 return;
 751         }
 752
 753         flist_del(&icmd->list);
 754         cmd->tag = icmd->saved_tag;
 755         free(icmd);
 756 }
 757
 758 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 759 {
 760         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 761         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 762
 763         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 764
 765         assert(client->eta_in_flight == eta);
 766
 767         client->eta_in_flight = NULL;
 768         flist_del_init(&client->eta_list);
 769
 770         convert_jobs_eta(je);
 771         sum_jobs_eta(&eta->eta, je);
 772         dec_jobs_eta(eta);
 773 }
 774
 775 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 776 {
 777         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 778         const char *os, *arch;
 779         char bit[16];
 780
 781         os = fio_get_os_string(probe->os);
 782         if (!os)
 783                 os = "unknown";
 784
 785         arch = fio_get_arch_string(probe->arch);
 786         if (!arch)
 787                 os = "unknown";
 788
 789         sprintf(bit, "%d-bit", probe->bpp * 8);
 790
 791         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
 792                 probe->hostname, probe->bigendian, bit, os, arch,
 793                 probe->fio_major, probe->fio_minor, probe->fio_patch);
 794
 795         if (!client->name)
 796                 client->name = strdup((char *) probe->hostname);
 797 }
 798
 799 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
 800 {
 801         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 802
 803         client->state = Client_started;
 804         client->jobs = le32_to_cpu(pdu->jobs);
 805 }
 806
 807 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
 808 {
 809         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
 810
 811         client->state = Client_stopped;
 812         client->error = le32_to_cpu(pdu->error);
 813
 814         if (client->error)
 815                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
 816 }
 817
 818 static int handle_client(struct fio_client *client)
 819 {
 820         struct fio_net_cmd *cmd;
 821
 822         dprint(FD_NET, "client: handle %s\n", client->hostname);
 823
 824         cmd = fio_net_recv_cmd(client->fd);
 825         if (!cmd)
 826                 return 0;
 827
 828         dprint(FD_NET, "client: got cmd op %s from %s\n",
 829                                 fio_server_op(cmd->opcode), client->hostname);
 830
 831         switch (cmd->opcode) {
 832         case FIO_NET_CMD_QUIT:
 833                 remove_client(client);
 834                 free(cmd);
 835                 break;
 836         case FIO_NET_CMD_TEXT: {
 837                 const char *buf = (const char *) cmd->payload;
 838                 const char *name;
 839                 int fio_unused ret;
 840
 841                 name = client->name ? client->name : client->hostname;
 842
 843                 if (!client->skip_newline)
 844                         fprintf(f_out, "<%s> ", name);
 845                 ret = fwrite(buf, cmd->pdu_len, 1, f_out);
 846                 fflush(f_out);
 847                 client->skip_newline = strchr(buf, '\n') == NULL;
 848                 free(cmd);
 849                 break;
 850                 }
 851         case FIO_NET_CMD_DU:
 852                 handle_du(client, cmd);
 853                 free(cmd);
 854                 break;
 855         case FIO_NET_CMD_TS:
 856                 handle_ts(cmd);
 857                 free(cmd);
 858                 break;
 859         case FIO_NET_CMD_GS:
 860                 handle_gs(cmd);
 861                 free(cmd);
 862                 break;
 863         case FIO_NET_CMD_ETA:
 864                 remove_reply_cmd(client, cmd);
 865                 handle_eta(client, cmd);
 866                 free(cmd);
 867                 break;
 868         case FIO_NET_CMD_PROBE:
 869                 remove_reply_cmd(client, cmd);
 870                 handle_probe(client, cmd);
 871                 free(cmd);
 872                 break;
 873         case FIO_NET_CMD_RUN:
 874                 client->state = Client_running;
 875                 free(cmd);
 876                 break;
 877         case FIO_NET_CMD_START:
 878                 handle_start(client, cmd);
 879                 free(cmd);
 880                 break;
 881         case FIO_NET_CMD_STOP:
 882                 handle_stop(client, cmd);
 883                 free(cmd);
 884                 break;
 885         default:
 886                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
 887                 free(cmd);
 888                 break;
 889         }
 890
 891         return 1;
 892 }
 893
 894 static void request_client_etas(void)
 895 {
 896         struct fio_client *client;
 897         struct flist_head *entry;
 898         struct client_eta *eta;
 899         int skipped = 0;
 900
 901         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
 902
 903         eta = malloc(sizeof(*eta));
 904         memset(&eta->eta, 0, sizeof(eta->eta));
 905         eta->pending = nr_clients;
 906
 907         flist_for_each(entry, &client_list) {
 908                 client = flist_entry(entry, struct fio_client, list);
 909
 910                 if (!flist_empty(&client->eta_list)) {
 911                         skipped++;
 912                         continue;
 913                 }
 914                 if (client->state != Client_running)
 915                         continue;
 916
 917                 assert(!client->eta_in_flight);
 918                 flist_add_tail(&client->eta_list, &eta_list);
 919                 client->eta_in_flight = eta;
 920                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
 921                                         (uintptr_t) eta, &client->cmd_list);
 922         }
 923
 924         while (skipped--)
 925                 dec_jobs_eta(eta);
 926
 927         dprint(FD_NET, "client: requested eta tag %p\n", eta);
 928 }
 929
 930 static int client_check_cmd_timeout(struct fio_client *client,
 931                                     struct timeval *now)
 932 {
 933         struct fio_net_int_cmd *cmd;
 934         struct flist_head *entry, *tmp;
 935         int ret = 0;
 936
 937         flist_for_each_safe(entry, tmp, &client->cmd_list) {
 938                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
 939
 940                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
 941                         continue;
 942
 943                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
 944                                                 fio_server_op(cmd->cmd.opcode));
 945                 flist_del(&cmd->list);
 946                 free(cmd);
 947                 ret = 1;
 948         }
 949
 950         return flist_empty(&client->cmd_list) && ret;
 951 }
 952
 953 static int fio_client_timed_out(void)
 954 {
 955         struct fio_client *client;
 956         struct flist_head *entry, *tmp;
 957         struct timeval tv;
 958         int ret = 0;
 959
 960         gettimeofday(&tv, NULL);
 961
 962         flist_for_each_safe(entry, tmp, &client_list) {
 963                 client = flist_entry(entry, struct fio_client, list);
 964
 965                 if (flist_empty(&client->cmd_list))
 966                         continue;
 967
 968                 if (!client_check_cmd_timeout(client, &tv))
 969                         continue;
 970
 971                 log_err("fio: client %s timed out\n", client->hostname);
 972                 remove_client(client);
 973                 ret = 1;
 974         }
 975
 976         return ret;
 977 }
 978
 979 int fio_handle_clients(void)
 980 {
 981         struct fio_client *client;
 982         struct flist_head *entry;
 983         struct pollfd *pfds;
 984         int i, ret = 0, retval = 0;
 985
 986         gettimeofday(&eta_tv, NULL);
 987
 988         pfds = malloc(nr_clients * sizeof(struct pollfd));
 989
 990         sum_stat_clients = nr_clients;
 991         init_thread_stat(&client_ts);
 992         init_group_run_stat(&client_gs);
 993
 994         while (!exit_backend && nr_clients) {
 995                 i = 0;
 996                 flist_for_each(entry, &client_list) {
 997                         client = flist_entry(entry, struct fio_client, list);
 998
 999                         pfds[i].fd = client->fd;
1000                         pfds[i].events = POLLIN;
1001                         i++;
1002                 }
1003
1004                 assert(i == nr_clients);
1005
1006                 do {
1007                         struct timeval tv;
1008
1009                         gettimeofday(&tv, NULL);
1010                         if (mtime_since(&eta_tv, &tv) >= 900) {
1011                                 request_client_etas();
1012                                 memcpy(&eta_tv, &tv, sizeof(tv));
1013
1014                                 if (fio_client_timed_out())
1015                                         break;
1016                         }
1017
1018                         ret = poll(pfds, nr_clients, 100);
1019                         if (ret < 0) {
1020                                 if (errno == EINTR)
1021                                         continue;
1022                                 log_err("fio: poll clients: %s\n", strerror(errno));
1023                                 break;
1024                         } else if (!ret)
1025                                 continue;
1026                 } while (ret <= 0);
1027
1028                 for (i = 0; i < nr_clients; i++) {
1029                         if (!(pfds[i].revents & POLLIN))
1030                                 continue;
1031
1032                         client = find_client_by_fd(pfds[i].fd);
1033                         if (!client) {
1034                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1035                                 continue;
1036                         }
1037                         if (!handle_client(client)) {
1038                                 log_info("client: host=%s disconnected\n",
1039                                                 client->hostname);
1040                                 remove_client(client);
1041                                 retval = 1;
1042                         } else if (client->error)
1043                                 retval = 1;
1044                 }
1045         }
1046
1047         free(pfds);
1048         return retval;
1049 }