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 #ifdef CONFIG_ZLIB
  18 #include <zlib.h>
  19 #endif
  20
  21 #include "fio.h"
  22 #include "client.h"
  23 #include "server.h"
  24 #include "flist.h"
  25 #include "hash.h"
  26
  27 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
  28 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd);
  29 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd);
  30 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
  31 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd);
  32 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd);
  33 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd);
  34
  35 struct client_ops fio_client_ops = {
  36         .text           = handle_text,
  37         .disk_util      = handle_du,
  38         .thread_status  = handle_ts,
  39         .group_stats    = handle_gs,
  40         .stop           = handle_stop,
  41         .start          = handle_start,
  42         .eta            = display_thread_status,
  43         .probe          = handle_probe,
  44         .eta_msec       = FIO_CLIENT_DEF_ETA_MSEC,
  45         .client_type    = FIO_CLIENT_TYPE_CLI,
  46 };
  47
  48 static struct timeval eta_tv;
  49
  50 static FLIST_HEAD(client_list);
  51 static FLIST_HEAD(eta_list);
  52
  53 static FLIST_HEAD(arg_list);
  54
  55 struct thread_stat client_ts;
  56 struct group_run_stats client_gs;
  57 int sum_stat_clients;
  58
  59 static int sum_stat_nr;
  60 static struct json_object *root = NULL;
  61 static struct json_array *clients_array = NULL;
  62 static struct json_array *du_array = NULL;
  63 static int do_output_all_clients;
  64
  65 #define FIO_CLIENT_HASH_BITS    7
  66 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  67 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  68 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  69
  70 static void fio_client_add_hash(struct fio_client *client)
  71 {
  72         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  73
  74         bucket &= FIO_CLIENT_HASH_MASK;
  75         flist_add(&client->hash_list, &client_hash[bucket]);
  76 }
  77
  78 static void fio_client_remove_hash(struct fio_client *client)
  79 {
  80         if (!flist_empty(&client->hash_list))
  81                 flist_del_init(&client->hash_list);
  82 }
  83
  84 static void fio_init fio_client_hash_init(void)
  85 {
  86         int i;
  87
  88         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
  89                 INIT_FLIST_HEAD(&client_hash[i]);
  90 }
  91
  92 static void fio_client_json_init(void)
  93 {
  94         if (output_format != FIO_OUTPUT_JSON)
  95                 return;
  96         root = json_create_object();
  97         json_object_add_value_string(root, "fio version", fio_version_string);
  98         clients_array = json_create_array();
  99         json_object_add_value_array(root, "client_stats", clients_array);
 100         du_array = json_create_array();
 101         json_object_add_value_array(root, "disk_util", du_array);
 102 }
 103
 104 static void fio_client_json_fini(void)
 105 {
 106         if (output_format != FIO_OUTPUT_JSON)
 107                 return;
 108         json_print_object(root);
 109         log_info("\n");
 110         json_free_object(root);
 111         root = NULL;
 112         clients_array = NULL;
 113         du_array = NULL;
 114 }
 115
 116 static struct fio_client *find_client_by_fd(int fd)
 117 {
 118         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
 119         struct fio_client *client;
 120         struct flist_head *entry;
 121
 122         flist_for_each(entry, &client_hash[bucket]) {
 123                 client = flist_entry(entry, struct fio_client, hash_list);
 124
 125                 if (client->fd == fd) {
 126                         client->refs++;
 127                         return client;
 128                 }
 129         }
 130
 131         return NULL;
 132 }
 133
 134 void fio_put_client(struct fio_client *client)
 135 {
 136         if (--client->refs)
 137                 return;
 138
 139         free(client->hostname);
 140         if (client->argv)
 141                 free(client->argv);
 142         if (client->name)
 143                 free(client->name);
 144         while (client->nr_files) {
 145                 struct client_file *cf = &client->files[--client->nr_files];
 146
 147                 free(cf->file);
 148         }
 149         if (client->files)
 150                 free(client->files);
 151
 152         if (!client->did_stat)
 153                 sum_stat_clients -= client->nr_stat;
 154
 155         free(client);
 156 }
 157
 158 static void remove_client(struct fio_client *client)
 159 {
 160         assert(client->refs);
 161
 162         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 163
 164         if (!flist_empty(&client->list))
 165                 flist_del_init(&client->list);
 166
 167         fio_client_remove_hash(client);
 168
 169         if (!flist_empty(&client->eta_list)) {
 170                 flist_del_init(&client->eta_list);
 171                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
 172         }
 173
 174         close(client->fd);
 175         client->fd = -1;
 176
 177         if (client->ops->removed)
 178                 client->ops->removed(client);
 179
 180         nr_clients--;
 181         fio_put_client(client);
 182 }
 183
 184 struct fio_client *fio_get_client(struct fio_client *client)
 185 {
 186         client->refs++;
 187         return client;
 188 }
 189
 190 static void __fio_client_add_cmd_option(struct fio_client *client,
 191                                         const char *opt)
 192 {
 193         int index;
 194
 195         index = client->argc++;
 196         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 197         client->argv[index] = strdup(opt);
 198         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 199 }
 200
 201 void fio_client_add_cmd_option(void *cookie, const char *opt)
 202 {
 203         struct fio_client *client = cookie;
 204         struct flist_head *entry;
 205
 206         if (!client || !opt)
 207                 return;
 208
 209         __fio_client_add_cmd_option(client, opt);
 210
 211         /*
 212          * Duplicate arguments to shared client group
 213          */
 214         flist_for_each(entry, &arg_list) {
 215                 client = flist_entry(entry, struct fio_client, arg_list);
 216
 217                 __fio_client_add_cmd_option(client, opt);
 218         }
 219 }
 220
 221 struct fio_client *fio_client_add_explicit(struct client_ops *ops,
 222                                            const char *hostname, int type,
 223                                            int port)
 224 {
 225         struct fio_client *client;
 226
 227         client = malloc(sizeof(*client));
 228         memset(client, 0, sizeof(*client));
 229
 230         INIT_FLIST_HEAD(&client->list);
 231         INIT_FLIST_HEAD(&client->hash_list);
 232         INIT_FLIST_HEAD(&client->arg_list);
 233         INIT_FLIST_HEAD(&client->eta_list);
 234         INIT_FLIST_HEAD(&client->cmd_list);
 235
 236         client->hostname = strdup(hostname);
 237
 238         if (type == Fio_client_socket)
 239                 client->is_sock = 1;
 240         else {
 241                 int ipv6;
 242
 243                 ipv6 = type == Fio_client_ipv6;
 244                 if (fio_server_parse_host(hostname, ipv6,
 245                                                 &client->addr.sin_addr,
 246                                                 &client->addr6.sin6_addr))
 247                         goto err;
 248
 249                 client->port = port;
 250         }
 251
 252         client->fd = -1;
 253         client->ops = ops;
 254         client->refs = 1;
 255         client->type = ops->client_type;
 256
 257         __fio_client_add_cmd_option(client, "fio");
 258
 259         flist_add(&client->list, &client_list);
 260         nr_clients++;
 261         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 262         return client;
 263 err:
 264         free(client);
 265         return NULL;
 266 }
 267
 268 int fio_client_add_ini_file(void *cookie, const char *ini_file, int remote)
 269 {
 270         struct fio_client *client = cookie;
 271         struct client_file *cf;
 272         size_t new_size;
 273         void *new_files;
 274
 275         dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file);
 276
 277         new_size = (client->nr_files + 1) * sizeof(struct client_file);
 278         new_files = realloc(client->files, new_size);
 279         if (!new_files)
 280                 return 1;
 281
 282         client->files = new_files;
 283         cf = &client->files[client->nr_files];
 284         cf->file = strdup(ini_file);
 285         cf->remote = remote;
 286         client->nr_files++;
 287         return 0;
 288 }
 289
 290 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
 291 {
 292         struct fio_client *existing = *cookie;
 293         struct fio_client *client;
 294
 295         if (existing) {
 296                 /*
 297                  * We always add our "exec" name as the option, hence 1
 298                  * means empty.
 299                  */
 300                 if (existing->argc == 1)
 301                         flist_add_tail(&existing->arg_list, &arg_list);
 302                 else {
 303                         while (!flist_empty(&arg_list))
 304                                 flist_del_init(arg_list.next);
 305                 }
 306         }
 307
 308         client = malloc(sizeof(*client));
 309         memset(client, 0, sizeof(*client));
 310
 311         INIT_FLIST_HEAD(&client->list);
 312         INIT_FLIST_HEAD(&client->hash_list);
 313         INIT_FLIST_HEAD(&client->arg_list);
 314         INIT_FLIST_HEAD(&client->eta_list);
 315         INIT_FLIST_HEAD(&client->cmd_list);
 316
 317         if (fio_server_parse_string(hostname, &client->hostname,
 318                                         &client->is_sock, &client->port,
 319                                         &client->addr.sin_addr,
 320                                         &client->addr6.sin6_addr,
 321                                         &client->ipv6))
 322                 return -1;
 323
 324         client->fd = -1;
 325         client->ops = ops;
 326         client->refs = 1;
 327         client->type = ops->client_type;
 328
 329         __fio_client_add_cmd_option(client, "fio");
 330
 331         flist_add(&client->list, &client_list);
 332         nr_clients++;
 333         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 334         *cookie = client;
 335         return 0;
 336 }
 337
 338 static void probe_client(struct fio_client *client)
 339 {
 340         struct cmd_client_probe_pdu pdu;
 341         uint64_t tag;
 342
 343         dprint(FD_NET, "client: send probe\n");
 344
 345 #ifdef CONFIG_ZLIB
 346         pdu.flags = __le64_to_cpu(FIO_PROBE_FLAG_ZLIB);
 347 #else
 348         pdu.flags = 0;
 349 #endif
 350
 351         fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list);
 352 }
 353
 354 static int fio_client_connect_ip(struct fio_client *client)
 355 {
 356         struct sockaddr *addr;
 357         socklen_t socklen;
 358         int fd, domain;
 359
 360         if (client->ipv6) {
 361                 client->addr6.sin6_family = AF_INET6;
 362                 client->addr6.sin6_port = htons(client->port);
 363                 domain = AF_INET6;
 364                 addr = (struct sockaddr *) &client->addr6;
 365                 socklen = sizeof(client->addr6);
 366         } else {
 367                 client->addr.sin_family = AF_INET;
 368                 client->addr.sin_port = htons(client->port);
 369                 domain = AF_INET;
 370                 addr = (struct sockaddr *) &client->addr;
 371                 socklen = sizeof(client->addr);
 372         }
 373
 374         fd = socket(domain, SOCK_STREAM, 0);
 375         if (fd < 0) {
 376                 int ret = -errno;
 377
 378                 log_err("fio: socket: %s\n", strerror(errno));
 379                 return ret;
 380         }
 381
 382         if (connect(fd, addr, socklen) < 0) {
 383                 int ret = -errno;
 384
 385                 log_err("fio: connect: %s\n", strerror(errno));
 386                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 387                                                                 client->port);
 388                 close(fd);
 389                 return ret;
 390         }
 391
 392         return fd;
 393 }
 394
 395 static int fio_client_connect_sock(struct fio_client *client)
 396 {
 397         struct sockaddr_un *addr = &client->addr_un;
 398         socklen_t len;
 399         int fd;
 400
 401         memset(addr, 0, sizeof(*addr));
 402         addr->sun_family = AF_UNIX;
 403         strncpy(addr->sun_path, client->hostname, sizeof(addr->sun_path) - 1);
 404
 405         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 406         if (fd < 0) {
 407                 int ret = -errno;
 408
 409                 log_err("fio: socket: %s\n", strerror(errno));
 410                 return ret;
 411         }
 412
 413         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 414         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 415                 int ret = -errno;
 416
 417                 log_err("fio: connect; %s\n", strerror(errno));
 418                 close(fd);
 419                 return ret;
 420         }
 421
 422         return fd;
 423 }
 424
 425 int fio_client_connect(struct fio_client *client)
 426 {
 427         int fd;
 428
 429         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 430
 431         if (client->is_sock)
 432                 fd = fio_client_connect_sock(client);
 433         else
 434                 fd = fio_client_connect_ip(client);
 435
 436         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 437
 438         if (fd < 0)
 439                 return fd;
 440
 441         client->fd = fd;
 442         fio_client_add_hash(client);
 443         client->state = Client_connected;
 444
 445         probe_client(client);
 446         return 0;
 447 }
 448
 449 int fio_client_terminate(struct fio_client *client)
 450 {
 451         return fio_net_send_quit(client->fd);
 452 }
 453
 454 void fio_clients_terminate(void)
 455 {
 456         struct flist_head *entry;
 457         struct fio_client *client;
 458
 459         dprint(FD_NET, "client: terminate clients\n");
 460
 461         flist_for_each(entry, &client_list) {
 462                 client = flist_entry(entry, struct fio_client, list);
 463                 fio_client_terminate(client);
 464         }
 465 }
 466
 467 static void sig_int(int sig)
 468 {
 469         dprint(FD_NET, "client: got signal %d\n", sig);
 470         fio_clients_terminate();
 471 }
 472
 473 static void sig_show_status(int sig)
 474 {
 475         show_running_run_stats();
 476 }
 477
 478 static void client_signal_handler(void)
 479 {
 480         struct sigaction act;
 481
 482         memset(&act, 0, sizeof(act));
 483         act.sa_handler = sig_int;
 484         act.sa_flags = SA_RESTART;
 485         sigaction(SIGINT, &act, NULL);
 486
 487         memset(&act, 0, sizeof(act));
 488         act.sa_handler = sig_int;
 489         act.sa_flags = SA_RESTART;
 490         sigaction(SIGTERM, &act, NULL);
 491
 492 /* Windows uses SIGBREAK as a quit signal from other applications */
 493 #ifdef WIN32
 494         memset(&act, 0, sizeof(act));
 495         act.sa_handler = sig_int;
 496         act.sa_flags = SA_RESTART;
 497         sigaction(SIGBREAK, &act, NULL);
 498 #endif
 499
 500         memset(&act, 0, sizeof(act));
 501         act.sa_handler = sig_show_status;
 502         act.sa_flags = SA_RESTART;
 503         sigaction(SIGUSR1, &act, NULL);
 504 }
 505
 506 static int send_client_cmd_line(struct fio_client *client)
 507 {
 508         struct cmd_single_line_pdu *cslp;
 509         struct cmd_line_pdu *clp;
 510         unsigned long offset;
 511         unsigned int *lens;
 512         void *pdu;
 513         size_t mem;
 514         int i, ret;
 515
 516         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 517
 518         lens = malloc(client->argc * sizeof(unsigned int));
 519
 520         /*
 521          * Find out how much mem we need
 522          */
 523         for (i = 0, mem = 0; i < client->argc; i++) {
 524                 lens[i] = strlen(client->argv[i]) + 1;
 525                 mem += lens[i];
 526         }
 527
 528         /*
 529          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 530          */
 531         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 532
 533         pdu = malloc(mem);
 534         clp = pdu;
 535         offset = sizeof(*clp);
 536
 537         for (i = 0; i < client->argc; i++) {
 538                 uint16_t arg_len = lens[i];
 539
 540                 cslp = pdu + offset;
 541                 strcpy((char *) cslp->text, client->argv[i]);
 542                 cslp->len = cpu_to_le16(arg_len);
 543                 offset += sizeof(*cslp) + arg_len;
 544         }
 545
 546         free(lens);
 547         clp->lines = cpu_to_le16(client->argc);
 548         clp->client_type = __cpu_to_le16(client->type);
 549         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, NULL, NULL);
 550         free(pdu);
 551         return ret;
 552 }
 553
 554 int fio_clients_connect(void)
 555 {
 556         struct fio_client *client;
 557         struct flist_head *entry, *tmp;
 558         int ret;
 559
 560 #ifdef WIN32
 561         WSADATA wsd;
 562         WSAStartup(MAKEWORD(2, 2), &wsd);
 563 #endif
 564
 565         dprint(FD_NET, "client: connect all\n");
 566
 567         client_signal_handler();
 568
 569         flist_for_each_safe(entry, tmp, &client_list) {
 570                 client = flist_entry(entry, struct fio_client, list);
 571
 572                 ret = fio_client_connect(client);
 573                 if (ret) {
 574                         remove_client(client);
 575                         continue;
 576                 }
 577
 578                 if (client->argc > 1)
 579                         send_client_cmd_line(client);
 580         }
 581
 582         return !nr_clients;
 583 }
 584
 585 int fio_start_client(struct fio_client *client)
 586 {
 587         dprint(FD_NET, "client: start %s\n", client->hostname);
 588         return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
 589 }
 590
 591 int fio_start_all_clients(void)
 592 {
 593         struct fio_client *client;
 594         struct flist_head *entry, *tmp;
 595         int ret;
 596
 597         dprint(FD_NET, "client: start all\n");
 598
 599         fio_client_json_init();
 600
 601         flist_for_each_safe(entry, tmp, &client_list) {
 602                 client = flist_entry(entry, struct fio_client, list);
 603
 604                 ret = fio_start_client(client);
 605                 if (ret) {
 606                         remove_client(client);
 607                         continue;
 608                 }
 609         }
 610
 611         return flist_empty(&client_list);
 612 }
 613
 614 static int __fio_client_send_remote_ini(struct fio_client *client,
 615                                         const char *filename)
 616 {
 617         struct cmd_load_file_pdu *pdu;
 618         size_t p_size;
 619         int ret;
 620
 621         dprint(FD_NET, "send remote ini %s to %s\n", filename, client->hostname);
 622
 623         p_size = sizeof(*pdu) + strlen(filename);
 624         pdu = malloc(p_size);
 625         pdu->name_len = strlen(filename);
 626         strcpy((char *) pdu->file, filename);
 627         pdu->client_type = cpu_to_le16((uint16_t) client->type);
 628
 629         client->sent_job = 1;
 630         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_LOAD_FILE, pdu, p_size,NULL, NULL);
 631         free(pdu);
 632         return ret;
 633 }
 634
 635 /*
 636  * Send file contents to server backend. We could use sendfile(), but to remain
 637  * more portable lets just read/write the darn thing.
 638  */
 639 static int __fio_client_send_local_ini(struct fio_client *client,
 640                                        const char *filename)
 641 {
 642         struct cmd_job_pdu *pdu;
 643         size_t p_size;
 644         struct stat sb;
 645         char *p;
 646         void *buf;
 647         off_t len;
 648         int fd, ret;
 649
 650         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 651
 652         fd = open(filename, O_RDONLY);
 653         if (fd < 0) {
 654                 int ret = -errno;
 655
 656                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 657                 return ret;
 658         }
 659
 660         if (fstat(fd, &sb) < 0) {
 661                 int ret = -errno;
 662
 663                 log_err("fio: job file stat: %s\n", strerror(errno));
 664                 close(fd);
 665                 return ret;
 666         }
 667
 668         p_size = sb.st_size + sizeof(*pdu);
 669         pdu = malloc(p_size);
 670         buf = pdu->buf;
 671
 672         len = sb.st_size;
 673         p = buf;
 674         do {
 675                 ret = read(fd, p, len);
 676                 if (ret > 0) {
 677                         len -= ret;
 678                         if (!len)
 679                                 break;
 680                         p += ret;
 681                         continue;
 682                 } else if (!ret)
 683                         break;
 684                 else if (errno == EAGAIN || errno == EINTR)
 685                         continue;
 686         } while (1);
 687
 688         if (len) {
 689                 log_err("fio: failed reading job file %s\n", filename);
 690                 close(fd);
 691                 free(pdu);
 692                 return 1;
 693         }
 694
 695         pdu->buf_len = __cpu_to_le32(sb.st_size);
 696         pdu->client_type = cpu_to_le32(client->type);
 697
 698         client->sent_job = 1;
 699         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, NULL, NULL);
 700         free(pdu);
 701         close(fd);
 702         return ret;
 703 }
 704
 705 int fio_client_send_ini(struct fio_client *client, const char *filename,
 706                         int remote)
 707 {
 708         int ret;
 709
 710         if (!remote)
 711                 ret = __fio_client_send_local_ini(client, filename);
 712         else
 713                 ret = __fio_client_send_remote_ini(client, filename);
 714
 715         if (!ret)
 716                 client->sent_job = 1;
 717
 718         return ret;
 719 }
 720
 721 static int fio_client_send_cf(struct fio_client *client,
 722                               struct client_file *cf)
 723 {
 724         return fio_client_send_ini(client, cf->file, cf->remote);
 725 }
 726
 727 int fio_clients_send_ini(const char *filename)
 728 {
 729         struct fio_client *client;
 730         struct flist_head *entry, *tmp;
 731
 732         flist_for_each_safe(entry, tmp, &client_list) {
 733                 client = flist_entry(entry, struct fio_client, list);
 734
 735                 if (client->nr_files) {
 736                         int i;
 737
 738                         for (i = 0; i < client->nr_files; i++) {
 739                                 struct client_file *cf;
 740
 741                                 cf = &client->files[i];
 742
 743                                 if (fio_client_send_cf(client, cf)) {
 744                                         remove_client(client);
 745                                         break;
 746                                 }
 747                         }
 748                 }
 749                 if (client->sent_job)
 750                         continue;
 751                 if (!filename || fio_client_send_ini(client, filename, 0))
 752                         remove_client(client);
 753         }
 754
 755         return !nr_clients;
 756 }
 757
 758 int fio_client_update_options(struct fio_client *client,
 759                               struct thread_options *o, uint64_t *tag)
 760 {
 761         struct cmd_add_job_pdu pdu;
 762
 763         pdu.thread_number = cpu_to_le32(client->thread_number);
 764         pdu.groupid = cpu_to_le32(client->groupid);
 765         convert_thread_options_to_net(&pdu.top, o);
 766
 767         return fio_net_send_cmd(client->fd, FIO_NET_CMD_UPDATE_JOB, &pdu, sizeof(pdu), tag, &client->cmd_list);
 768 }
 769
 770 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 771 {
 772         dst->max_val    = le64_to_cpu(src->max_val);
 773         dst->min_val    = le64_to_cpu(src->min_val);
 774         dst->samples    = le64_to_cpu(src->samples);
 775
 776         /*
 777          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 778          */
 779         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 780         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 781 }
 782
 783 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 784 {
 785         int i, j;
 786
 787         dst->error              = le32_to_cpu(src->error);
 788         dst->thread_number      = le32_to_cpu(src->thread_number);
 789         dst->groupid            = le32_to_cpu(src->groupid);
 790         dst->pid                = le32_to_cpu(src->pid);
 791         dst->members            = le32_to_cpu(src->members);
 792         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
 793
 794         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 795                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 796                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 797                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 798                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 799         }
 800
 801         dst->usr_time           = le64_to_cpu(src->usr_time);
 802         dst->sys_time           = le64_to_cpu(src->sys_time);
 803         dst->ctx                = le64_to_cpu(src->ctx);
 804         dst->minf               = le64_to_cpu(src->minf);
 805         dst->majf               = le64_to_cpu(src->majf);
 806         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 807
 808         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 809                 fio_fp64_t *fps = &src->percentile_list[i];
 810                 fio_fp64_t *fpd = &dst->percentile_list[i];
 811
 812                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 813         }
 814
 815         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 816                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 817                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 818                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 819         }
 820
 821         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 822                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 823                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 824         }
 825
 826         for (i = 0; i < DDIR_RWDIR_CNT; i++)
 827                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 828                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 829
 830         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 831                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 832                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 833         }
 834
 835         dst->total_submit       = le64_to_cpu(src->total_submit);
 836         dst->total_complete     = le64_to_cpu(src->total_complete);
 837
 838         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 839                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 840                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 841         }
 842
 843         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 844         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 845         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 846         dst->first_error        = le32_to_cpu(src->first_error);
 847         dst->kb_base            = le32_to_cpu(src->kb_base);
 848         dst->unit_base          = le32_to_cpu(src->unit_base);
 849
 850         dst->latency_depth      = le32_to_cpu(src->latency_depth);
 851         dst->latency_target     = le64_to_cpu(src->latency_target);
 852         dst->latency_window     = le64_to_cpu(src->latency_window);
 853         dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
 854 }
 855
 856 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 857 {
 858         int i;
 859
 860         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 861                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 862                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 863                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 864                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 865                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 866                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 867         }
 868
 869         dst->kb_base    = le32_to_cpu(src->kb_base);
 870         dst->unit_base  = le32_to_cpu(src->unit_base);
 871         dst->groupid    = le32_to_cpu(src->groupid);
 872         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
 873 }
 874
 875 static void json_object_add_client_info(struct json_object *obj,
 876                                         struct fio_client *client)
 877 {
 878         const char *hostname = client->hostname ? client->hostname : "";
 879
 880         json_object_add_value_string(obj, "hostname", hostname);
 881         json_object_add_value_int(obj, "port", client->port);
 882 }
 883
 884 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
 885 {
 886         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 887         struct json_object *tsobj;
 888
 889         tsobj = show_thread_status(&p->ts, &p->rs);
 890         client->did_stat = 1;
 891         if (tsobj) {
 892                 json_object_add_client_info(tsobj, client);
 893                 json_array_add_value_object(clients_array, tsobj);
 894         }
 895
 896         if (!do_output_all_clients)
 897                 return;
 898
 899         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 900         sum_group_stats(&client_gs, &p->rs);
 901
 902         client_ts.members++;
 903         client_ts.thread_number = p->ts.thread_number;
 904         client_ts.groupid = p->ts.groupid;
 905         client_ts.unified_rw_rep = p->ts.unified_rw_rep;
 906
 907         if (++sum_stat_nr == sum_stat_clients) {
 908                 strcpy(client_ts.name, "All clients");
 909                 tsobj = show_thread_status(&client_ts, &client_gs);
 910                 if (tsobj) {
 911                         json_object_add_client_info(tsobj, client);
 912                         json_array_add_value_object(clients_array, tsobj);
 913                 }
 914         }
 915 }
 916
 917 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
 918 {
 919         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 920
 921         show_group_stats(gs);
 922 }
 923
 924 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
 925 {
 926         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
 927         const char *buf = (const char *) pdu->buf;
 928         const char *name;
 929         int fio_unused ret;
 930
 931         name = client->name ? client->name : client->hostname;
 932
 933         if (!client->skip_newline)
 934                 fprintf(f_out, "<%s> ", name);
 935         ret = fwrite(buf, pdu->buf_len, 1, f_out);
 936         fflush(f_out);
 937         client->skip_newline = strchr(buf, '\n') == NULL;
 938 }
 939
 940 static void convert_agg(struct disk_util_agg *agg)
 941 {
 942         int i;
 943
 944         for (i = 0; i < 2; i++) {
 945                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 946                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 947                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 948                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 949         }
 950
 951         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 952         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 953         agg->slavecount         = le32_to_cpu(agg->slavecount);
 954         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 955 }
 956
 957 static void convert_dus(struct disk_util_stat *dus)
 958 {
 959         int i;
 960
 961         for (i = 0; i < 2; i++) {
 962                 dus->s.ios[i]           = le32_to_cpu(dus->s.ios[i]);
 963                 dus->s.merges[i]        = le32_to_cpu(dus->s.merges[i]);
 964                 dus->s.sectors[i]       = le64_to_cpu(dus->s.sectors[i]);
 965                 dus->s.ticks[i]         = le32_to_cpu(dus->s.ticks[i]);
 966         }
 967
 968         dus->s.io_ticks         = le32_to_cpu(dus->s.io_ticks);
 969         dus->s.time_in_queue    = le32_to_cpu(dus->s.time_in_queue);
 970         dus->s.msec             = le64_to_cpu(dus->s.msec);
 971 }
 972
 973 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 974 {
 975         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 976
 977         if (!client->disk_stats_shown) {
 978                 client->disk_stats_shown = 1;
 979                 log_info("\nDisk stats (read/write):\n");
 980         }
 981
 982         if (output_format == FIO_OUTPUT_JSON) {
 983                 struct json_object *duobj;
 984                 json_array_add_disk_util(&du->dus, &du->agg, du_array);
 985                 duobj = json_array_last_value_object(du_array);
 986                 json_object_add_client_info(duobj, client);
 987         } else
 988                 print_disk_util(&du->dus, &du->agg, output_format == FIO_OUTPUT_TERSE);
 989 }
 990
 991 static void convert_jobs_eta(struct jobs_eta *je)
 992 {
 993         int i;
 994
 995         je->nr_running          = le32_to_cpu(je->nr_running);
 996         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 997         je->nr_pending          = le32_to_cpu(je->nr_pending);
 998         je->nr_setting_up       = le32_to_cpu(je->nr_setting_up);
 999         je->files_open          = le32_to_cpu(je->files_open);
1000
1001         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1002                 je->m_rate[i]   = le32_to_cpu(je->m_rate[i]);
1003                 je->t_rate[i]   = le32_to_cpu(je->t_rate[i]);
1004                 je->m_iops[i]   = le32_to_cpu(je->m_iops[i]);
1005                 je->t_iops[i]   = le32_to_cpu(je->t_iops[i]);
1006                 je->rate[i]     = le32_to_cpu(je->rate[i]);
1007                 je->iops[i]     = le32_to_cpu(je->iops[i]);
1008         }
1009
1010         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
1011         je->eta_sec             = le64_to_cpu(je->eta_sec);
1012         je->nr_threads          = le32_to_cpu(je->nr_threads);
1013         je->is_pow2             = le32_to_cpu(je->is_pow2);
1014         je->unit_base           = le32_to_cpu(je->unit_base);
1015 }
1016
1017 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
1018 {
1019         int i;
1020
1021         dst->nr_running         += je->nr_running;
1022         dst->nr_ramp            += je->nr_ramp;
1023         dst->nr_pending         += je->nr_pending;
1024         dst->nr_setting_up      += je->nr_setting_up;
1025         dst->files_open         += je->files_open;
1026
1027         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1028                 dst->m_rate[i]  += je->m_rate[i];
1029                 dst->t_rate[i]  += je->t_rate[i];
1030                 dst->m_iops[i]  += je->m_iops[i];
1031                 dst->t_iops[i]  += je->t_iops[i];
1032                 dst->rate[i]    += je->rate[i];
1033                 dst->iops[i]    += je->iops[i];
1034         }
1035
1036         dst->elapsed_sec        += je->elapsed_sec;
1037
1038         if (je->eta_sec > dst->eta_sec)
1039                 dst->eta_sec = je->eta_sec;
1040
1041         dst->nr_threads         += je->nr_threads;
1042
1043         /*
1044          * This wont be correct for multiple strings, but at least it
1045          * works for the basic cases.
1046          */
1047         strcpy((char *) dst->run_str, (char *) je->run_str);
1048 }
1049
1050 void fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
1051 {
1052         if (!--eta->pending) {
1053                 eta_fn(&eta->eta);
1054                 free(eta);
1055         }
1056 }
1057
1058 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
1059 {
1060         struct fio_net_cmd_reply *reply = NULL;
1061         struct flist_head *entry;
1062
1063         flist_for_each(entry, &client->cmd_list) {
1064                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1065
1066                 if (cmd->tag == (uintptr_t) reply)
1067                         break;
1068
1069                 reply = NULL;
1070         }
1071
1072         if (!reply) {
1073                 log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag);
1074                 return;
1075         }
1076
1077         flist_del(&reply->list);
1078         cmd->tag = reply->saved_tag;
1079         free(reply);
1080 }
1081
1082 int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
1083 {
1084         do {
1085                 struct fio_net_cmd_reply *reply = NULL;
1086                 struct flist_head *entry;
1087
1088                 flist_for_each(entry, &client->cmd_list) {
1089                         reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1090
1091                         if (tag == (uintptr_t) reply)
1092                                 break;
1093
1094                         reply = NULL;
1095                 }
1096
1097                 if (!reply)
1098                         break;
1099
1100                 usleep(1000);
1101         } while (1);
1102
1103         return 0;
1104 }
1105
1106 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
1107 {
1108         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1109         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
1110
1111         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
1112
1113         assert(client->eta_in_flight == eta);
1114
1115         client->eta_in_flight = NULL;
1116         flist_del_init(&client->eta_list);
1117
1118         if (client->ops->jobs_eta)
1119                 client->ops->jobs_eta(client, je);
1120
1121         fio_client_sum_jobs_eta(&eta->eta, je);
1122         fio_client_dec_jobs_eta(eta, client->ops->eta);
1123 }
1124
1125 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
1126 {
1127         struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
1128         const char *os, *arch;
1129         char bit[16];
1130
1131         os = fio_get_os_string(probe->os);
1132         if (!os)
1133                 os = "unknown";
1134
1135         arch = fio_get_arch_string(probe->arch);
1136         if (!arch)
1137                 os = "unknown";
1138
1139         sprintf(bit, "%d-bit", probe->bpp * 8);
1140         probe->flags = le64_to_cpu(probe->flags);
1141
1142         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n",
1143                 probe->hostname, probe->bigendian, bit, os, arch,
1144                 probe->fio_version, (unsigned long) probe->flags);
1145
1146         if (!client->name)
1147                 client->name = strdup((char *) probe->hostname);
1148 }
1149
1150 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
1151 {
1152         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1153
1154         client->state = Client_started;
1155         client->jobs = le32_to_cpu(pdu->jobs);
1156         client->nr_stat = le32_to_cpu(pdu->stat_outputs);
1157
1158         if (sum_stat_clients > 1)
1159                 do_output_all_clients = 1;
1160
1161         sum_stat_clients += client->nr_stat;
1162 }
1163
1164 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
1165 {
1166         if (client->error)
1167                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
1168 }
1169
1170 static void convert_stop(struct fio_net_cmd *cmd)
1171 {
1172         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1173
1174         pdu->error = le32_to_cpu(pdu->error);
1175 }
1176
1177 static void convert_text(struct fio_net_cmd *cmd)
1178 {
1179         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1180
1181         pdu->level      = le32_to_cpu(pdu->level);
1182         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
1183         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
1184         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
1185 }
1186
1187 static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
1188                                               struct cmd_iolog_pdu *pdu)
1189 {
1190 #ifdef CONFIG_ZLIB
1191         struct cmd_iolog_pdu *ret;
1192         z_stream stream;
1193         uint32_t nr_samples;
1194         size_t total;
1195         void *p;
1196
1197         stream.zalloc = Z_NULL;
1198         stream.zfree = Z_NULL;
1199         stream.opaque = Z_NULL;
1200         stream.avail_in = 0;
1201         stream.next_in = Z_NULL;
1202
1203         if (inflateInit(&stream) != Z_OK)
1204                 return NULL;
1205
1206         /*
1207          * Get header first, it's not compressed
1208          */
1209         nr_samples = le64_to_cpu(pdu->nr_samples);
1210
1211         total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
1212         ret = malloc(total + sizeof(*pdu));
1213         ret->nr_samples = nr_samples;
1214
1215         memcpy(ret, pdu, sizeof(*pdu));
1216
1217         p = (void *) ret + sizeof(*pdu);
1218
1219         stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1220         stream.next_in = (void *) pdu + sizeof(*pdu);
1221         while (stream.avail_in) {
1222                 unsigned int this_chunk = 65536;
1223                 unsigned int this_len;
1224                 int err;
1225
1226                 if (this_chunk > total)
1227                         this_chunk = total;
1228
1229                 stream.avail_out = this_chunk;
1230                 stream.next_out = p;
1231                 err = inflate(&stream, Z_NO_FLUSH);
1232                 /* may be Z_OK, or Z_STREAM_END */
1233                 if (err < 0) {
1234                         log_err("fio: inflate error %d\n", err);
1235                         free(ret);
1236                         ret = NULL;
1237                         goto err;
1238                 }
1239
1240                 this_len = this_chunk - stream.avail_out;
1241                 p += this_len;
1242                 total -= this_len;
1243         }
1244
1245 err:
1246         inflateEnd(&stream);
1247         return ret;
1248 #else
1249         return NULL;
1250 #endif
1251 }
1252
1253 /*
1254  * This has been compressed on the server side, since it can be big.
1255  * Uncompress here.
1256  */
1257 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd)
1258 {
1259         struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1260         struct cmd_iolog_pdu *ret;
1261         uint64_t i;
1262         void *samples;
1263
1264         /*
1265          * Convert if compressed and we support it. If it's not
1266          * compressed, we need not do anything.
1267          */
1268         if (le32_to_cpu(pdu->compressed)) {
1269 #ifndef CONFIG_ZLIB
1270                 log_err("fio: server sent compressed data by mistake\n");
1271                 return NULL;
1272 #endif
1273                 ret = convert_iolog_gz(cmd, pdu);
1274                 if (!ret) {
1275                         log_err("fio: failed decompressing log\n");
1276                         return NULL;
1277                 }
1278         } else
1279                 ret = pdu;
1280
1281         ret->nr_samples         = le64_to_cpu(ret->nr_samples);
1282         ret->thread_number      = le32_to_cpu(ret->thread_number);
1283         ret->log_type           = le32_to_cpu(ret->log_type);
1284         ret->compressed         = le32_to_cpu(ret->compressed);
1285         ret->log_offset         = le32_to_cpu(ret->log_offset);
1286
1287         samples = &ret->samples[0];
1288         for (i = 0; i < ret->nr_samples; i++) {
1289                 struct io_sample *s;
1290
1291                 s = __get_sample(samples, ret->log_offset, i);
1292                 s->time         = le64_to_cpu(s->time);
1293                 s->val          = le64_to_cpu(s->val);
1294                 s->__ddir       = le32_to_cpu(s->__ddir);
1295                 s->bs           = le32_to_cpu(s->bs);
1296
1297                 if (ret->log_offset) {
1298                         struct io_sample_offset *so = (void *) s;
1299
1300                         so->offset = le64_to_cpu(so->offset);
1301                 }
1302         }
1303
1304         return ret;
1305 }
1306
1307 int fio_handle_client(struct fio_client *client)
1308 {
1309         struct client_ops *ops = client->ops;
1310         struct fio_net_cmd *cmd;
1311
1312         dprint(FD_NET, "client: handle %s\n", client->hostname);
1313
1314         cmd = fio_net_recv_cmd(client->fd);
1315         if (!cmd)
1316                 return 0;
1317
1318         dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1319                 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1320
1321         switch (cmd->opcode) {
1322         case FIO_NET_CMD_QUIT:
1323                 if (ops->quit)
1324                         ops->quit(client, cmd);
1325                 remove_client(client);
1326                 free(cmd);
1327                 break;
1328         case FIO_NET_CMD_TEXT:
1329                 convert_text(cmd);
1330                 ops->text(client, cmd);
1331                 free(cmd);
1332                 break;
1333         case FIO_NET_CMD_DU: {
1334                 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1335
1336                 convert_dus(&du->dus);
1337                 convert_agg(&du->agg);
1338
1339                 ops->disk_util(client, cmd);
1340                 free(cmd);
1341                 break;
1342                 }
1343         case FIO_NET_CMD_TS: {
1344                 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1345
1346                 convert_ts(&p->ts, &p->ts);
1347                 convert_gs(&p->rs, &p->rs);
1348
1349                 ops->thread_status(client, cmd);
1350                 free(cmd);
1351                 break;
1352                 }
1353         case FIO_NET_CMD_GS: {
1354                 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1355
1356                 convert_gs(gs, gs);
1357
1358                 ops->group_stats(client, cmd);
1359                 free(cmd);
1360                 break;
1361                 }
1362         case FIO_NET_CMD_ETA: {
1363                 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1364
1365                 remove_reply_cmd(client, cmd);
1366                 convert_jobs_eta(je);
1367                 handle_eta(client, cmd);
1368                 free(cmd);
1369                 break;
1370                 }
1371         case FIO_NET_CMD_PROBE:
1372                 remove_reply_cmd(client, cmd);
1373                 ops->probe(client, cmd);
1374                 free(cmd);
1375                 break;
1376         case FIO_NET_CMD_SERVER_START:
1377                 client->state = Client_running;
1378                 if (ops->job_start)
1379                         ops->job_start(client, cmd);
1380                 free(cmd);
1381                 break;
1382         case FIO_NET_CMD_START: {
1383                 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1384
1385                 pdu->jobs = le32_to_cpu(pdu->jobs);
1386                 ops->start(client, cmd);
1387                 free(cmd);
1388                 break;
1389                 }
1390         case FIO_NET_CMD_STOP: {
1391                 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1392
1393                 convert_stop(cmd);
1394                 client->state = Client_stopped;
1395                 client->error = le32_to_cpu(pdu->error);
1396                 client->signal = le32_to_cpu(pdu->signal);
1397                 ops->stop(client, cmd);
1398                 free(cmd);
1399                 break;
1400                 }
1401         case FIO_NET_CMD_ADD_JOB: {
1402                 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1403
1404                 client->thread_number = le32_to_cpu(pdu->thread_number);
1405                 client->groupid = le32_to_cpu(pdu->groupid);
1406
1407                 if (ops->add_job)
1408                         ops->add_job(client, cmd);
1409                 free(cmd);
1410                 break;
1411                 }
1412         case FIO_NET_CMD_IOLOG:
1413                 if (ops->iolog) {
1414                         struct cmd_iolog_pdu *pdu;
1415
1416                         pdu = convert_iolog(cmd);
1417                         ops->iolog(client, pdu);
1418                 }
1419                 free(cmd);
1420                 break;
1421         case FIO_NET_CMD_UPDATE_JOB:
1422                 ops->update_job(client, cmd);
1423                 remove_reply_cmd(client, cmd);
1424                 free(cmd);
1425                 break;
1426         default:
1427                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1428                 free(cmd);
1429                 break;
1430         }
1431
1432         return 1;
1433 }
1434
1435 static void request_client_etas(struct client_ops *ops)
1436 {
1437         struct fio_client *client;
1438         struct flist_head *entry;
1439         struct client_eta *eta;
1440         int skipped = 0;
1441
1442         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1443
1444         eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
1445         eta->pending = nr_clients;
1446
1447         flist_for_each(entry, &client_list) {
1448                 client = flist_entry(entry, struct fio_client, list);
1449
1450                 if (!flist_empty(&client->eta_list)) {
1451                         skipped++;
1452                         continue;
1453                 }
1454                 if (client->state != Client_running)
1455                         continue;
1456
1457                 assert(!client->eta_in_flight);
1458                 flist_add_tail(&client->eta_list, &eta_list);
1459                 client->eta_in_flight = eta;
1460                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1461                                         (uintptr_t) eta, &client->cmd_list);
1462         }
1463
1464         while (skipped--)
1465                 fio_client_dec_jobs_eta(eta, ops->eta);
1466
1467         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1468 }
1469
1470 static int client_check_cmd_timeout(struct fio_client *client,
1471                                     struct timeval *now)
1472 {
1473         struct fio_net_cmd_reply *reply;
1474         struct flist_head *entry, *tmp;
1475         int ret = 0;
1476
1477         flist_for_each_safe(entry, tmp, &client->cmd_list) {
1478                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1479
1480                 if (mtime_since(&reply->tv, now) < FIO_NET_CLIENT_TIMEOUT)
1481                         continue;
1482
1483                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1484                                                 fio_server_op(reply->opcode));
1485                 flist_del(&reply->list);
1486                 free(reply);
1487                 ret = 1;
1488         }
1489
1490         return flist_empty(&client->cmd_list) && ret;
1491 }
1492
1493 static int fio_check_clients_timed_out(void)
1494 {
1495         struct fio_client *client;
1496         struct flist_head *entry, *tmp;
1497         struct timeval tv;
1498         int ret = 0;
1499
1500         fio_gettime(&tv, NULL);
1501
1502         flist_for_each_safe(entry, tmp, &client_list) {
1503                 client = flist_entry(entry, struct fio_client, list);
1504
1505                 if (flist_empty(&client->cmd_list))
1506                         continue;
1507
1508                 if (!client_check_cmd_timeout(client, &tv))
1509                         continue;
1510
1511                 if (client->ops->timed_out)
1512                         client->ops->timed_out(client);
1513                 else
1514                         log_err("fio: client %s timed out\n", client->hostname);
1515
1516                 remove_client(client);
1517                 ret = 1;
1518         }
1519
1520         return ret;
1521 }
1522
1523 int fio_handle_clients(struct client_ops *ops)
1524 {
1525         struct pollfd *pfds;
1526         int i, ret = 0, retval = 0;
1527
1528         fio_gettime(&eta_tv, NULL);
1529
1530         pfds = malloc(nr_clients * sizeof(struct pollfd));
1531
1532         init_thread_stat(&client_ts);
1533         init_group_run_stat(&client_gs);
1534
1535         while (!exit_backend && nr_clients) {
1536                 struct flist_head *entry, *tmp;
1537                 struct fio_client *client;
1538
1539                 i = 0;
1540                 flist_for_each_safe(entry, tmp, &client_list) {
1541                         client = flist_entry(entry, struct fio_client, list);
1542
1543                         if (!client->sent_job && !client->ops->stay_connected &&
1544                             flist_empty(&client->cmd_list)) {
1545                                 remove_client(client);
1546                                 continue;
1547                         }
1548
1549                         pfds[i].fd = client->fd;
1550                         pfds[i].events = POLLIN;
1551                         i++;
1552                 }
1553
1554                 if (!nr_clients)
1555                         break;
1556
1557                 assert(i == nr_clients);
1558
1559                 do {
1560                         struct timeval tv;
1561
1562                         fio_gettime(&tv, NULL);
1563                         if (mtime_since(&eta_tv, &tv) >= 900) {
1564                                 request_client_etas(ops);
1565                                 memcpy(&eta_tv, &tv, sizeof(tv));
1566
1567                                 if (fio_check_clients_timed_out())
1568                                         break;
1569                         }
1570
1571                         ret = poll(pfds, nr_clients, ops->eta_msec);
1572                         if (ret < 0) {
1573                                 if (errno == EINTR)
1574                                         continue;
1575                                 log_err("fio: poll clients: %s\n", strerror(errno));
1576                                 break;
1577                         } else if (!ret)
1578                                 continue;
1579                 } while (ret <= 0);
1580
1581                 for (i = 0; i < nr_clients; i++) {
1582                         if (!(pfds[i].revents & POLLIN))
1583                                 continue;
1584
1585                         client = find_client_by_fd(pfds[i].fd);
1586                         if (!client) {
1587                                 log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd);
1588                                 continue;
1589                         }
1590                         if (!fio_handle_client(client)) {
1591                                 log_info("client: host=%s disconnected\n",
1592                                                 client->hostname);
1593                                 remove_client(client);
1594                                 retval = 1;
1595                         } else if (client->error)
1596                                 retval = 1;
1597                         fio_put_client(client);
1598                 }
1599         }
1600
1601         fio_client_json_fini();
1602
1603         free(pfds);
1604         return retval;
1605 }