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         unsigned int refs;
  41
  42         char *name;
  43
  44         int state;
  45
  46         int skip_newline;
  47         int is_sock;
  48         int disk_stats_shown;
  49         unsigned int jobs;
  50         int error;
  51         int ipv6;
  52         int sent_job;
  53
  54         struct flist_head eta_list;
  55         struct client_eta *eta_in_flight;
  56
  57         struct flist_head cmd_list;
  58
  59         uint16_t argc;
  60         char **argv;
  61 };
  62
  63 static struct timeval eta_tv;
  64
  65 enum {
  66         Client_created          = 0,
  67         Client_connected        = 1,
  68         Client_started          = 2,
  69         Client_running          = 3,
  70         Client_stopped          = 4,
  71         Client_exited           = 5,
  72 };
  73
  74 static FLIST_HEAD(client_list);
  75 static FLIST_HEAD(eta_list);
  76
  77 static FLIST_HEAD(arg_list);
  78
  79 static struct thread_stat client_ts;
  80 static struct group_run_stats client_gs;
  81 static int sum_stat_clients;
  82 static int sum_stat_nr;
  83
  84 #define FIO_CLIENT_HASH_BITS    7
  85 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  86 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  87 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  88
  89 static int handle_client(struct fio_client *client);
  90 static void dec_jobs_eta(struct client_eta *eta);
  91
  92 static void fio_client_add_hash(struct fio_client *client)
  93 {
  94         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  95
  96         bucket &= FIO_CLIENT_HASH_MASK;
  97         flist_add(&client->hash_list, &client_hash[bucket]);
  98 }
  99
 100 static void fio_client_remove_hash(struct fio_client *client)
 101 {
 102         if (!flist_empty(&client->hash_list))
 103                 flist_del_init(&client->hash_list);
 104 }
 105
 106 static void fio_init fio_client_hash_init(void)
 107 {
 108         int i;
 109
 110         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
 111                 INIT_FLIST_HEAD(&client_hash[i]);
 112 }
 113
 114 static struct fio_client *find_client_by_fd(int fd)
 115 {
 116         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
 117         struct fio_client *client;
 118         struct flist_head *entry;
 119
 120         flist_for_each(entry, &client_hash[bucket]) {
 121                 client = flist_entry(entry, struct fio_client, hash_list);
 122
 123                 if (client->fd == fd) {
 124                         client->refs++;
 125                         return client;
 126                 }
 127         }
 128
 129         return NULL;
 130 }
 131
 132 static void remove_client(struct fio_client *client)
 133 {
 134         assert(client->refs);
 135
 136         if (--client->refs)
 137                 return;
 138
 139         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 140         flist_del(&client->list);
 141
 142         fio_client_remove_hash(client);
 143
 144         if (!flist_empty(&client->eta_list)) {
 145                 flist_del_init(&client->eta_list);
 146                 dec_jobs_eta(client->eta_in_flight);
 147         }
 148
 149         free(client->hostname);
 150         if (client->argv)
 151                 free(client->argv);
 152         if (client->name)
 153                 free(client->name);
 154
 155         free(client);
 156         nr_clients--;
 157         sum_stat_clients--;
 158 }
 159
 160 static void put_client(struct fio_client *client)
 161 {
 162         remove_client(client);
 163 }
 164
 165 static void __fio_client_add_cmd_option(struct fio_client *client,
 166                                         const char *opt)
 167 {
 168         int index;
 169
 170         index = client->argc++;
 171         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 172         client->argv[index] = strdup(opt);
 173         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 174 }
 175
 176 void fio_client_add_cmd_option(void *cookie, const char *opt)
 177 {
 178         struct fio_client *client = cookie;
 179         struct flist_head *entry;
 180
 181         if (!client || !opt)
 182                 return;
 183
 184         __fio_client_add_cmd_option(client, opt);
 185
 186         /*
 187          * Duplicate arguments to shared client group
 188          */
 189         flist_for_each(entry, &arg_list) {
 190                 client = flist_entry(entry, struct fio_client, arg_list);
 191
 192                 __fio_client_add_cmd_option(client, opt);
 193         }
 194 }
 195
 196 int fio_client_add(const char *hostname, void **cookie)
 197 {
 198         struct fio_client *existing = *cookie;
 199         struct fio_client *client;
 200
 201         if (existing) {
 202                 /*
 203                  * We always add our "exec" name as the option, hence 1
 204                  * means empty.
 205                  */
 206                 if (existing->argc == 1)
 207                         flist_add_tail(&existing->arg_list, &arg_list);
 208                 else {
 209                         while (!flist_empty(&arg_list))
 210                                 flist_del_init(arg_list.next);
 211                 }
 212         }
 213
 214         client = malloc(sizeof(*client));
 215         memset(client, 0, sizeof(*client));
 216
 217         INIT_FLIST_HEAD(&client->list);
 218         INIT_FLIST_HEAD(&client->hash_list);
 219         INIT_FLIST_HEAD(&client->arg_list);
 220         INIT_FLIST_HEAD(&client->eta_list);
 221         INIT_FLIST_HEAD(&client->cmd_list);
 222
 223         if (fio_server_parse_string(hostname, &client->hostname,
 224                                         &client->is_sock, &client->port,
 225                                         &client->addr.sin_addr,
 226                                         &client->addr6.sin6_addr,
 227                                         &client->ipv6))
 228                 return -1;
 229
 230         client->fd = -1;
 231         client->refs = 1;
 232
 233         __fio_client_add_cmd_option(client, "fio");
 234
 235         flist_add(&client->list, &client_list);
 236         nr_clients++;
 237         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 238         *cookie = client;
 239         return 0;
 240 }
 241
 242 static int fio_client_connect_ip(struct fio_client *client)
 243 {
 244         struct sockaddr *addr;
 245         fio_socklen_t socklen;
 246         int fd, domain;
 247
 248         if (client->ipv6) {
 249                 client->addr6.sin6_family = AF_INET6;
 250                 client->addr6.sin6_port = htons(client->port);
 251                 domain = AF_INET6;
 252                 addr = (struct sockaddr *) &client->addr6;
 253                 socklen = sizeof(client->addr6);
 254         } else {
 255                 client->addr.sin_family = AF_INET;
 256                 client->addr.sin_port = htons(client->port);
 257                 domain = AF_INET;
 258                 addr = (struct sockaddr *) &client->addr;
 259                 socklen = sizeof(client->addr);
 260         }
 261
 262         fd = socket(domain, SOCK_STREAM, 0);
 263         if (fd < 0) {
 264                 log_err("fio: socket: %s\n", strerror(errno));
 265                 return -1;
 266         }
 267
 268         if (connect(fd, addr, socklen) < 0) {
 269                 log_err("fio: connect: %s\n", strerror(errno));
 270                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 271                                                                 client->port);
 272                 close(fd);
 273                 return -1;
 274         }
 275
 276         return fd;
 277 }
 278
 279 static int fio_client_connect_sock(struct fio_client *client)
 280 {
 281         struct sockaddr_un *addr = &client->addr_un;
 282         fio_socklen_t len;
 283         int fd;
 284
 285         memset(addr, 0, sizeof(*addr));
 286         addr->sun_family = AF_UNIX;
 287         strcpy(addr->sun_path, client->hostname);
 288
 289         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 290         if (fd < 0) {
 291                 log_err("fio: socket: %s\n", strerror(errno));
 292                 return -1;
 293         }
 294
 295         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 296         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 297                 log_err("fio: connect; %s\n", strerror(errno));
 298                 close(fd);
 299                 return -1;
 300         }
 301
 302         return fd;
 303 }
 304
 305 static int fio_client_connect(struct fio_client *client)
 306 {
 307         int fd;
 308
 309         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 310
 311         if (client->is_sock)
 312                 fd = fio_client_connect_sock(client);
 313         else
 314                 fd = fio_client_connect_ip(client);
 315
 316         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 317
 318         if (fd < 0)
 319                 return 1;
 320
 321         client->fd = fd;
 322         fio_client_add_hash(client);
 323         client->state = Client_connected;
 324         return 0;
 325 }
 326
 327 void fio_clients_terminate(void)
 328 {
 329         struct flist_head *entry;
 330         struct fio_client *client;
 331
 332         dprint(FD_NET, "client: terminate clients\n");
 333
 334         flist_for_each(entry, &client_list) {
 335                 client = flist_entry(entry, struct fio_client, list);
 336
 337                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0, NULL);
 338         }
 339 }
 340
 341 static void sig_int(int sig)
 342 {
 343         dprint(FD_NET, "client: got signal %d\n", sig);
 344         fio_clients_terminate();
 345 }
 346
 347 static void sig_show_status(int sig)
 348 {
 349         show_running_run_stats();
 350 }
 351
 352 static void client_signal_handler(void)
 353 {
 354         struct sigaction act;
 355
 356         memset(&act, 0, sizeof(act));
 357         act.sa_handler = sig_int;
 358         act.sa_flags = SA_RESTART;
 359         sigaction(SIGINT, &act, NULL);
 360
 361         memset(&act, 0, sizeof(act));
 362         act.sa_handler = sig_int;
 363         act.sa_flags = SA_RESTART;
 364         sigaction(SIGTERM, &act, NULL);
 365
 366         memset(&act, 0, sizeof(act));
 367         act.sa_handler = sig_show_status;
 368         act.sa_flags = SA_RESTART;
 369         sigaction(SIGUSR1, &act, NULL);
 370 }
 371
 372 static void probe_client(struct fio_client *client)
 373 {
 374         dprint(FD_NET, "client: send probe\n");
 375
 376         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
 377 }
 378
 379 static int send_client_cmd_line(struct fio_client *client)
 380 {
 381         struct cmd_single_line_pdu *cslp;
 382         struct cmd_line_pdu *clp;
 383         unsigned long offset;
 384         unsigned int *lens;
 385         void *pdu;
 386         size_t mem;
 387         int i, ret;
 388
 389         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 390
 391         lens = malloc(client->argc * sizeof(unsigned int));
 392
 393         /*
 394          * Find out how much mem we need
 395          */
 396         for (i = 0, mem = 0; i < client->argc; i++) {
 397                 lens[i] = strlen(client->argv[i]) + 1;
 398                 mem += lens[i];
 399         }
 400
 401         /*
 402          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 403          */
 404         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 405
 406         pdu = malloc(mem);
 407         clp = pdu;
 408         offset = sizeof(*clp);
 409
 410         for (i = 0; i < client->argc; i++) {
 411                 uint16_t arg_len = lens[i];
 412
 413                 cslp = pdu + offset;
 414                 strcpy((char *) cslp->text, client->argv[i]);
 415                 cslp->len = cpu_to_le16(arg_len);
 416                 offset += sizeof(*cslp) + arg_len;
 417         }
 418
 419         free(lens);
 420         clp->lines = cpu_to_le16(client->argc);
 421         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
 422         free(pdu);
 423         return ret;
 424 }
 425
 426 int fio_clients_connect(void)
 427 {
 428         struct fio_client *client;
 429         struct flist_head *entry, *tmp;
 430         int ret;
 431
 432 #ifdef WIN32
 433         WSADATA wsd;
 434         WSAStartup(MAKEWORD(2,2), &wsd);
 435 #endif
 436
 437         dprint(FD_NET, "client: connect all\n");
 438
 439         client_signal_handler();
 440
 441         flist_for_each_safe(entry, tmp, &client_list) {
 442                 client = flist_entry(entry, struct fio_client, list);
 443
 444                 ret = fio_client_connect(client);
 445                 if (ret) {
 446                         remove_client(client);
 447                         continue;
 448                 }
 449
 450                 probe_client(client);
 451
 452                 if (client->argc > 1)
 453                         send_client_cmd_line(client);
 454         }
 455
 456         return !nr_clients;
 457 }
 458
 459 /*
 460  * Send file contents to server backend. We could use sendfile(), but to remain
 461  * more portable lets just read/write the darn thing.
 462  */
 463 static int fio_client_send_ini(struct fio_client *client, const char *filename)
 464 {
 465         struct stat sb;
 466         char *p, *buf;
 467         off_t len;
 468         int fd, ret;
 469
 470         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 471
 472         fd = open(filename, O_RDONLY);
 473         if (fd < 0) {
 474                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 475                 return 1;
 476         }
 477
 478         if (fstat(fd, &sb) < 0) {
 479                 log_err("fio: job file stat: %s\n", strerror(errno));
 480                 close(fd);
 481                 return 1;
 482         }
 483
 484         buf = malloc(sb.st_size);
 485
 486         len = sb.st_size;
 487         p = buf;
 488         do {
 489                 ret = read(fd, p, len);
 490                 if (ret > 0) {
 491                         len -= ret;
 492                         if (!len)
 493                                 break;
 494                         p += ret;
 495                         continue;
 496                 } else if (!ret)
 497                         break;
 498                 else if (errno == EAGAIN || errno == EINTR)
 499                         continue;
 500         } while (1);
 501
 502         if (len) {
 503                 log_err("fio: failed reading job file %s\n", filename);
 504                 close(fd);
 505                 free(buf);
 506                 return 1;
 507         }
 508
 509         client->sent_job = 1;
 510         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size, 0);
 511         free(buf);
 512         close(fd);
 513         return ret;
 514 }
 515
 516 int fio_clients_send_ini(const char *filename)
 517 {
 518         struct fio_client *client;
 519         struct flist_head *entry, *tmp;
 520
 521         flist_for_each_safe(entry, tmp, &client_list) {
 522                 client = flist_entry(entry, struct fio_client, list);
 523
 524                 if (fio_client_send_ini(client, filename))
 525                         remove_client(client);
 526
 527                 client->sent_job = 1;
 528         }
 529
 530         return !nr_clients;
 531 }
 532
 533 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 534 {
 535         dst->max_val    = le64_to_cpu(src->max_val);
 536         dst->min_val    = le64_to_cpu(src->min_val);
 537         dst->samples    = le64_to_cpu(src->samples);
 538
 539         /*
 540          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 541          */
 542         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 543         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 544 }
 545
 546 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 547 {
 548         int i, j;
 549
 550         dst->error      = le32_to_cpu(src->error);
 551         dst->groupid    = le32_to_cpu(src->groupid);
 552         dst->pid        = le32_to_cpu(src->pid);
 553         dst->members    = le32_to_cpu(src->members);
 554
 555         for (i = 0; i < 2; i++) {
 556                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 557                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 558                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 559                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 560         }
 561
 562         dst->usr_time           = le64_to_cpu(src->usr_time);
 563         dst->sys_time           = le64_to_cpu(src->sys_time);
 564         dst->ctx                = le64_to_cpu(src->ctx);
 565         dst->minf               = le64_to_cpu(src->minf);
 566         dst->majf               = le64_to_cpu(src->majf);
 567         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 568
 569         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 570                 fio_fp64_t *fps = &src->percentile_list[i];
 571                 fio_fp64_t *fpd = &dst->percentile_list[i];
 572
 573                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 574         }
 575
 576         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 577                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 578                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 579                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 580         }
 581
 582         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 583                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 584                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 585         }
 586
 587         for (i = 0; i < 2; i++)
 588                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 589                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 590
 591         for (i = 0; i < 3; i++) {
 592                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 593                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 594         }
 595
 596         dst->total_submit       = le64_to_cpu(src->total_submit);
 597         dst->total_complete     = le64_to_cpu(src->total_complete);
 598
 599         for (i = 0; i < 2; i++) {
 600                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 601                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 602         }
 603
 604         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 605         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 606         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 607         dst->first_error        = le32_to_cpu(src->first_error);
 608         dst->kb_base            = le32_to_cpu(src->kb_base);
 609 }
 610
 611 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 612 {
 613         int i;
 614
 615         for (i = 0; i < 2; i++) {
 616                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 617                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 618                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 619                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 620                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 621                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 622         }
 623
 624         dst->kb_base    = le32_to_cpu(src->kb_base);
 625         dst->groupid    = le32_to_cpu(src->groupid);
 626 }
 627
 628 static void handle_ts(struct fio_net_cmd *cmd)
 629 {
 630         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 631
 632         convert_ts(&p->ts, &p->ts);
 633         convert_gs(&p->rs, &p->rs);
 634
 635         show_thread_status(&p->ts, &p->rs);
 636
 637         if (sum_stat_clients == 1)
 638                 return;
 639
 640         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 641         sum_group_stats(&client_gs, &p->rs);
 642
 643         client_ts.members++;
 644         client_ts.groupid = p->ts.groupid;
 645
 646         if (++sum_stat_nr == sum_stat_clients) {
 647                 strcpy(client_ts.name, "All clients");
 648                 show_thread_status(&client_ts, &client_gs);
 649         }
 650 }
 651
 652 static void handle_gs(struct fio_net_cmd *cmd)
 653 {
 654         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 655
 656         convert_gs(gs, gs);
 657         show_group_stats(gs);
 658 }
 659
 660 static void convert_agg(struct disk_util_agg *agg)
 661 {
 662         int i;
 663
 664         for (i = 0; i < 2; i++) {
 665                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 666                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 667                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 668                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 669         }
 670
 671         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 672         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 673         agg->slavecount         = le32_to_cpu(agg->slavecount);
 674         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 675 }
 676
 677 static void convert_dus(struct disk_util_stat *dus)
 678 {
 679         int i;
 680
 681         for (i = 0; i < 2; i++) {
 682                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
 683                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
 684                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
 685                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
 686         }
 687
 688         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
 689         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
 690         dus->msec               = le64_to_cpu(dus->msec);
 691 }
 692
 693 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 694 {
 695         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 696
 697         convert_dus(&du->dus);
 698         convert_agg(&du->agg);
 699
 700         if (!client->disk_stats_shown) {
 701                 client->disk_stats_shown = 1;
 702                 log_info("\nDisk stats (read/write):\n");
 703         }
 704
 705         print_disk_util(&du->dus, &du->agg, terse_output);
 706 }
 707
 708 static void convert_jobs_eta(struct jobs_eta *je)
 709 {
 710         int i;
 711
 712         je->nr_running          = le32_to_cpu(je->nr_running);
 713         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 714         je->nr_pending          = le32_to_cpu(je->nr_pending);
 715         je->files_open          = le32_to_cpu(je->files_open);
 716         je->m_rate              = le32_to_cpu(je->m_rate);
 717         je->t_rate              = le32_to_cpu(je->t_rate);
 718         je->m_iops              = le32_to_cpu(je->m_iops);
 719         je->t_iops              = le32_to_cpu(je->t_iops);
 720
 721         for (i = 0; i < 2; i++) {
 722                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 723                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 724         }
 725
 726         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 727         je->eta_sec             = le64_to_cpu(je->eta_sec);
 728 }
 729
 730 static void sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 731 {
 732         int i;
 733
 734         dst->nr_running         += je->nr_running;
 735         dst->nr_ramp            += je->nr_ramp;
 736         dst->nr_pending         += je->nr_pending;
 737         dst->files_open         += je->files_open;
 738         dst->m_rate             += je->m_rate;
 739         dst->t_rate             += je->t_rate;
 740         dst->m_iops             += je->m_iops;
 741         dst->t_iops             += je->t_iops;
 742
 743         for (i = 0; i < 2; i++) {
 744                 dst->rate[i]    += je->rate[i];
 745                 dst->iops[i]    += je->iops[i];
 746         }
 747
 748         dst->elapsed_sec        += je->elapsed_sec;
 749
 750         if (je->eta_sec > dst->eta_sec)
 751                 dst->eta_sec = je->eta_sec;
 752 }
 753
 754 static void dec_jobs_eta(struct client_eta *eta)
 755 {
 756         if (!--eta->pending) {
 757                 display_thread_status(&eta->eta);
 758                 free(eta);
 759         }
 760 }
 761
 762 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 763 {
 764         struct fio_net_int_cmd *icmd = NULL;
 765         struct flist_head *entry;
 766
 767         flist_for_each(entry, &client->cmd_list) {
 768                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
 769
 770                 if (cmd->tag == (uintptr_t) icmd)
 771                         break;
 772
 773                 icmd = NULL;
 774         }
 775
 776         if (!icmd) {
 777                 log_err("fio: client: unable to find matching tag\n");
 778                 return;
 779         }
 780
 781         flist_del(&icmd->list);
 782         cmd->tag = icmd->saved_tag;
 783         free(icmd);
 784 }
 785
 786 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 787 {
 788         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 789         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 790
 791         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 792
 793         assert(client->eta_in_flight == eta);
 794
 795         client->eta_in_flight = NULL;
 796         flist_del_init(&client->eta_list);
 797
 798         convert_jobs_eta(je);
 799         sum_jobs_eta(&eta->eta, je);
 800         dec_jobs_eta(eta);
 801 }
 802
 803 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 804 {
 805         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 806         const char *os, *arch;
 807         char bit[16];
 808
 809         os = fio_get_os_string(probe->os);
 810         if (!os)
 811                 os = "unknown";
 812
 813         arch = fio_get_arch_string(probe->arch);
 814         if (!arch)
 815                 os = "unknown";
 816
 817         sprintf(bit, "%d-bit", probe->bpp * 8);
 818
 819         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
 820                 probe->hostname, probe->bigendian, bit, os, arch,
 821                 probe->fio_major, probe->fio_minor, probe->fio_patch);
 822
 823         if (!client->name)
 824                 client->name = strdup((char *) probe->hostname);
 825 }
 826
 827 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
 828 {
 829         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 830
 831         client->state = Client_started;
 832         client->jobs = le32_to_cpu(pdu->jobs);
 833 }
 834
 835 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
 836 {
 837         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
 838
 839         client->state = Client_stopped;
 840         client->error = le32_to_cpu(pdu->error);
 841
 842         if (client->error)
 843                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
 844 }
 845
 846 static int handle_client(struct fio_client *client)
 847 {
 848         struct fio_net_cmd *cmd;
 849
 850         dprint(FD_NET, "client: handle %s\n", client->hostname);
 851
 852         cmd = fio_net_recv_cmd(client->fd);
 853         if (!cmd)
 854                 return 0;
 855
 856         dprint(FD_NET, "client: got cmd op %s from %s\n",
 857                                 fio_server_op(cmd->opcode), client->hostname);
 858
 859         switch (cmd->opcode) {
 860         case FIO_NET_CMD_QUIT:
 861                 remove_client(client);
 862                 free(cmd);
 863                 break;
 864         case FIO_NET_CMD_TEXT: {
 865                 const char *buf = (const char *) cmd->payload;
 866                 const char *name;
 867                 int fio_unused ret;
 868
 869                 name = client->name ? client->name : client->hostname;
 870
 871                 if (!client->skip_newline)
 872                         fprintf(f_out, "<%s> ", name);
 873                 ret = fwrite(buf, cmd->pdu_len, 1, f_out);
 874                 fflush(f_out);
 875                 client->skip_newline = strchr(buf, '\n') == NULL;
 876                 free(cmd);
 877                 break;
 878                 }
 879         case FIO_NET_CMD_DU:
 880                 handle_du(client, cmd);
 881                 free(cmd);
 882                 break;
 883         case FIO_NET_CMD_TS:
 884                 handle_ts(cmd);
 885                 free(cmd);
 886                 break;
 887         case FIO_NET_CMD_GS:
 888                 handle_gs(cmd);
 889                 free(cmd);
 890                 break;
 891         case FIO_NET_CMD_ETA:
 892                 remove_reply_cmd(client, cmd);
 893                 handle_eta(client, cmd);
 894                 free(cmd);
 895                 break;
 896         case FIO_NET_CMD_PROBE:
 897                 remove_reply_cmd(client, cmd);
 898                 handle_probe(client, cmd);
 899                 free(cmd);
 900                 break;
 901         case FIO_NET_CMD_RUN:
 902                 client->state = Client_running;
 903                 free(cmd);
 904                 break;
 905         case FIO_NET_CMD_START:
 906                 handle_start(client, cmd);
 907                 free(cmd);
 908                 break;
 909         case FIO_NET_CMD_STOP:
 910                 handle_stop(client, cmd);
 911                 free(cmd);
 912                 break;
 913         default:
 914                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
 915                 free(cmd);
 916                 break;
 917         }
 918
 919         return 1;
 920 }
 921
 922 static void request_client_etas(void)
 923 {
 924         struct fio_client *client;
 925         struct flist_head *entry;
 926         struct client_eta *eta;
 927         int skipped = 0;
 928
 929         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
 930
 931         eta = malloc(sizeof(*eta));
 932         memset(&eta->eta, 0, sizeof(eta->eta));
 933         eta->pending = nr_clients;
 934
 935         flist_for_each(entry, &client_list) {
 936                 client = flist_entry(entry, struct fio_client, list);
 937
 938                 if (!flist_empty(&client->eta_list)) {
 939                         skipped++;
 940                         continue;
 941                 }
 942                 if (client->state != Client_running)
 943                         continue;
 944
 945                 assert(!client->eta_in_flight);
 946                 flist_add_tail(&client->eta_list, &eta_list);
 947                 client->eta_in_flight = eta;
 948                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
 949                                         (uintptr_t) eta, &client->cmd_list);
 950         }
 951
 952         while (skipped--)
 953                 dec_jobs_eta(eta);
 954
 955         dprint(FD_NET, "client: requested eta tag %p\n", eta);
 956 }
 957
 958 static int client_check_cmd_timeout(struct fio_client *client,
 959                                     struct timeval *now)
 960 {
 961         struct fio_net_int_cmd *cmd;
 962         struct flist_head *entry, *tmp;
 963         int ret = 0;
 964
 965         flist_for_each_safe(entry, tmp, &client->cmd_list) {
 966                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
 967
 968                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
 969                         continue;
 970
 971                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
 972                                                 fio_server_op(cmd->cmd.opcode));
 973                 flist_del(&cmd->list);
 974                 free(cmd);
 975                 ret = 1;
 976         }
 977
 978         return flist_empty(&client->cmd_list) && ret;
 979 }
 980
 981 static int fio_client_timed_out(void)
 982 {
 983         struct fio_client *client;
 984         struct flist_head *entry, *tmp;
 985         struct timeval tv;
 986         int ret = 0;
 987
 988         gettimeofday(&tv, NULL);
 989
 990         flist_for_each_safe(entry, tmp, &client_list) {
 991                 client = flist_entry(entry, struct fio_client, list);
 992
 993                 if (flist_empty(&client->cmd_list))
 994                         continue;
 995
 996                 if (!client_check_cmd_timeout(client, &tv))
 997                         continue;
 998
 999                 log_err("fio: client %s timed out\n", client->hostname);
1000                 remove_client(client);
1001                 ret = 1;
1002         }
1003
1004         return ret;
1005 }
1006
1007 int fio_handle_clients(void)
1008 {
1009         struct pollfd *pfds;
1010         int i, ret = 0, retval = 0;
1011
1012         gettimeofday(&eta_tv, NULL);
1013
1014         pfds = malloc(nr_clients * sizeof(struct pollfd));
1015
1016         sum_stat_clients = nr_clients;
1017         init_thread_stat(&client_ts);
1018         init_group_run_stat(&client_gs);
1019
1020         while (!exit_backend && nr_clients) {
1021                 struct flist_head *entry, *tmp;
1022                 struct fio_client *client;
1023
1024                 i = 0;
1025                 flist_for_each_safe(entry, tmp, &client_list) {
1026                         client = flist_entry(entry, struct fio_client, list);
1027
1028                         if (!client->sent_job &&
1029                             flist_empty(&client->cmd_list)) {
1030                                 remove_client(client);
1031                                 continue;
1032                         }
1033
1034                         pfds[i].fd = client->fd;
1035                         pfds[i].events = POLLIN;
1036                         i++;
1037                 }
1038
1039                 if (!nr_clients)
1040                         break;
1041
1042                 assert(i == nr_clients);
1043
1044                 do {
1045                         struct timeval tv;
1046
1047                         gettimeofday(&tv, NULL);
1048                         if (mtime_since(&eta_tv, &tv) >= 900) {
1049                                 request_client_etas();
1050                                 memcpy(&eta_tv, &tv, sizeof(tv));
1051
1052                                 if (fio_client_timed_out())
1053                                         break;
1054                         }
1055
1056                         ret = poll(pfds, nr_clients, 100);
1057                         if (ret < 0) {
1058                                 if (errno == EINTR)
1059                                         continue;
1060                                 log_err("fio: poll clients: %s\n", strerror(errno));
1061                                 break;
1062                         } else if (!ret)
1063                                 continue;
1064                 } while (ret <= 0);
1065
1066                 for (i = 0; i < nr_clients; i++) {
1067                         if (!(pfds[i].revents & POLLIN))
1068                                 continue;
1069
1070                         client = find_client_by_fd(pfds[i].fd);
1071                         if (!client) {
1072                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1073                                 continue;
1074                         }
1075                         if (!handle_client(client)) {
1076                                 log_info("client: host=%s disconnected\n",
1077                                                 client->hostname);
1078                                 remove_client(client);
1079                                 retval = 1;
1080                         } else if (client->error)
1081                                 retval = 1;
1082                         put_client(client);
1083                 }
1084         }
1085
1086         free(pfds);
1087         return retval;
1088 }