client.c

   1 #include <stdio.h>
   2 #include <stdlib.h>
   3 #include <unistd.h>
   4 #include <errno.h>
   5 #include <fcntl.h>
   6 #include <poll.h>
   7 #include <sys/types.h>
   8 #include <sys/stat.h>
   9 #include <sys/socket.h>
  10 #include <sys/un.h>
  11 #include <netinet/in.h>
  12 #include <arpa/inet.h>
  13 #include <netdb.h>
  14 #include <signal.h>
  15 #ifdef CONFIG_ZLIB
  16 #include <zlib.h>
  17 #endif
  18
  19 #include "fio.h"
  20 #include "client.h"
  21 #include "server.h"
  22 #include "flist.h"
  23 #include "hash.h"
  24 #include "verify-state.h"
  25
  26 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
  27 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd);
  28 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd);
  29 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
  30 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd);
  31 static void handle_stop(struct fio_client *client);
  32 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd);
  33
  34 static void convert_text(struct fio_net_cmd *cmd);
  35
  36 struct client_ops fio_client_ops = {
  37         .text           = handle_text,
  38         .disk_util      = handle_du,
  39         .thread_status  = handle_ts,
  40         .group_stats    = handle_gs,
  41         .stop           = handle_stop,
  42         .start          = handle_start,
  43         .eta            = display_thread_status,
  44         .probe          = handle_probe,
  45         .eta_msec       = FIO_CLIENT_DEF_ETA_MSEC,
  46         .client_type    = FIO_CLIENT_TYPE_CLI,
  47 };
  48
  49 static struct timespec eta_ts;
  50
  51 static FLIST_HEAD(client_list);
  52 static FLIST_HEAD(eta_list);
  53
  54 static FLIST_HEAD(arg_list);
  55
  56 struct thread_stat client_ts;
  57 struct group_run_stats client_gs;
  58 int sum_stat_clients;
  59
  60 static int sum_stat_nr;
  61 static struct json_object *root = NULL;
  62 static struct json_object *job_opt_object = NULL;
  63 static struct json_array *clients_array = NULL;
  64 static struct json_array *du_array = NULL;
  65
  66 static int error_clients;
  67
  68 #define FIO_CLIENT_HASH_BITS    7
  69 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  70 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  71 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  72
  73 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *, bool *);
  74
  75 static void fio_client_add_hash(struct fio_client *client)
  76 {
  77         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  78
  79         bucket &= FIO_CLIENT_HASH_MASK;
  80         flist_add(&client->hash_list, &client_hash[bucket]);
  81 }
  82
  83 static void fio_client_remove_hash(struct fio_client *client)
  84 {
  85         if (!flist_empty(&client->hash_list))
  86                 flist_del_init(&client->hash_list);
  87 }
  88
  89 static void fio_init fio_client_hash_init(void)
  90 {
  91         int i;
  92
  93         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
  94                 INIT_FLIST_HEAD(&client_hash[i]);
  95 }
  96
  97 static int read_data(int fd, void *data, size_t size)
  98 {
  99         ssize_t ret;
 100
 101         while (size) {
 102                 ret = read(fd, data, size);
 103                 if (ret < 0) {
 104                         if (errno == EAGAIN || errno == EINTR)
 105                                 continue;
 106                         break;
 107                 } else if (!ret)
 108                         break;
 109                 else {
 110                         data += ret;
 111                         size -= ret;
 112                 }
 113         }
 114
 115         if (size)
 116                 return EAGAIN;
 117
 118         return 0;
 119 }
 120
 121 static int read_ini_data(int fd, void *data, size_t size)
 122 {
 123         char *p = data;
 124         int ret = 0;
 125         FILE *fp;
 126         int dupfd;
 127
 128         dupfd = dup(fd);
 129         if (dupfd < 0)
 130                 return errno;
 131
 132         fp = fdopen(dupfd, "r");
 133         if (!fp) {
 134                 ret = errno;
 135                 close(dupfd);
 136                 goto out;
 137         }
 138
 139         while (1) {
 140                 ssize_t len;
 141                 char buf[OPT_LEN_MAX+1], *sub;
 142
 143                 if (!fgets(buf, sizeof(buf), fp)) {
 144                         if (ferror(fp)) {
 145                                 if (errno == EAGAIN || errno == EINTR)
 146                                         continue;
 147                                 ret = errno;
 148                         }
 149                         break;
 150                 }
 151
 152                 sub = fio_option_dup_subs(buf);
 153                 len = strlen(sub);
 154                 if (len + 1 > size) {
 155                         log_err("fio: no space left to read data\n");
 156                         free(sub);
 157                         ret = ENOSPC;
 158                         break;
 159                 }
 160
 161                 memcpy(p, sub, len);
 162                 free(sub);
 163                 p += len;
 164                 *p = '\0';
 165                 size -= len;
 166         }
 167
 168         fclose(fp);
 169 out:
 170         return ret;
 171 }
 172
 173 static void fio_client_json_init(void)
 174 {
 175         char time_buf[32];
 176         time_t time_p;
 177
 178         if (!(output_format & FIO_OUTPUT_JSON))
 179                 return;
 180
 181         time(&time_p);
 182         os_ctime_r((const time_t *) &time_p, time_buf, sizeof(time_buf));
 183         time_buf[strlen(time_buf) - 1] = '\0';
 184
 185         root = json_create_object();
 186         json_object_add_value_string(root, "fio version", fio_version_string);
 187         json_object_add_value_int(root, "timestamp", time_p);
 188         json_object_add_value_string(root, "time", time_buf);
 189
 190         job_opt_object = json_create_object();
 191         json_object_add_value_object(root, "global options", job_opt_object);
 192         clients_array = json_create_array();
 193         json_object_add_value_array(root, "client_stats", clients_array);
 194         du_array = json_create_array();
 195         json_object_add_value_array(root, "disk_util", du_array);
 196 }
 197
 198 static void fio_client_json_fini(void)
 199 {
 200         if (!(output_format & FIO_OUTPUT_JSON))
 201                 return;
 202
 203         log_info("\n");
 204         json_print_object(root, NULL);
 205         log_info("\n");
 206         json_free_object(root);
 207         root = NULL;
 208         clients_array = NULL;
 209         du_array = NULL;
 210 }
 211
 212 static struct fio_client *find_client_by_fd(int fd)
 213 {
 214         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
 215         struct fio_client *client;
 216         struct flist_head *entry;
 217
 218         flist_for_each(entry, &client_hash[bucket]) {
 219                 client = flist_entry(entry, struct fio_client, hash_list);
 220
 221                 if (client->fd == fd) {
 222                         client->refs++;
 223                         return client;
 224                 }
 225         }
 226
 227         return NULL;
 228 }
 229
 230 void fio_put_client(struct fio_client *client)
 231 {
 232         if (--client->refs)
 233                 return;
 234
 235         free(client->hostname);
 236         if (client->argv)
 237                 free(client->argv);
 238         if (client->name)
 239                 free(client->name);
 240         while (client->nr_files) {
 241                 struct client_file *cf = &client->files[--client->nr_files];
 242
 243                 free(cf->file);
 244         }
 245         if (client->files)
 246                 free(client->files);
 247         if (client->opt_lists)
 248                 free(client->opt_lists);
 249
 250         if (!client->did_stat)
 251                 sum_stat_clients--;
 252
 253         if (client->error)
 254                 error_clients++;
 255
 256         free(client);
 257 }
 258
 259 static int fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
 260 {
 261         if (!--eta->pending) {
 262                 eta_fn(&eta->eta);
 263                 free(eta);
 264                 return 0;
 265         }
 266
 267         return 1;
 268 }
 269
 270 static void fio_drain_client_text(struct fio_client *client)
 271 {
 272         do {
 273                 struct fio_net_cmd *cmd;
 274
 275                 cmd = fio_net_recv_cmd(client->fd, false);
 276                 if (!cmd)
 277                         break;
 278
 279                 if (cmd->opcode == FIO_NET_CMD_TEXT) {
 280                         convert_text(cmd);
 281                         client->ops->text(client, cmd);
 282                 }
 283
 284                 free(cmd);
 285         } while (1);
 286 }
 287
 288 static void remove_client(struct fio_client *client)
 289 {
 290         assert(client->refs);
 291
 292         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 293
 294         fio_drain_client_text(client);
 295
 296         if (!flist_empty(&client->list))
 297                 flist_del_init(&client->list);
 298
 299         fio_client_remove_hash(client);
 300
 301         if (!flist_empty(&client->eta_list)) {
 302                 flist_del_init(&client->eta_list);
 303                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
 304         }
 305
 306         close(client->fd);
 307         client->fd = -1;
 308
 309         if (client->ops->removed)
 310                 client->ops->removed(client);
 311
 312         nr_clients--;
 313         fio_put_client(client);
 314 }
 315
 316 struct fio_client *fio_get_client(struct fio_client *client)
 317 {
 318         client->refs++;
 319         return client;
 320 }
 321
 322 static void __fio_client_add_cmd_option(struct fio_client *client,
 323                                         const char *opt)
 324 {
 325         int index;
 326
 327         index = client->argc++;
 328         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 329         client->argv[index] = strdup(opt);
 330         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 331 }
 332
 333 void fio_client_add_cmd_option(void *cookie, const char *opt)
 334 {
 335         struct fio_client *client = cookie;
 336         struct flist_head *entry;
 337
 338         if (!client || !opt)
 339                 return;
 340
 341         __fio_client_add_cmd_option(client, opt);
 342
 343         /*
 344          * Duplicate arguments to shared client group
 345          */
 346         flist_for_each(entry, &arg_list) {
 347                 client = flist_entry(entry, struct fio_client, arg_list);
 348
 349                 __fio_client_add_cmd_option(client, opt);
 350         }
 351 }
 352
 353 struct fio_client *fio_client_add_explicit(struct client_ops *ops,
 354                                            const char *hostname, int type,
 355                                            int port)
 356 {
 357         struct fio_client *client;
 358
 359         client = malloc(sizeof(*client));
 360         memset(client, 0, sizeof(*client));
 361
 362         INIT_FLIST_HEAD(&client->list);
 363         INIT_FLIST_HEAD(&client->hash_list);
 364         INIT_FLIST_HEAD(&client->arg_list);
 365         INIT_FLIST_HEAD(&client->eta_list);
 366         INIT_FLIST_HEAD(&client->cmd_list);
 367
 368         client->hostname = strdup(hostname);
 369
 370         if (type == Fio_client_socket)
 371                 client->is_sock = true;
 372         else {
 373                 int ipv6;
 374
 375                 ipv6 = type == Fio_client_ipv6;
 376                 if (fio_server_parse_host(hostname, ipv6,
 377                                                 &client->addr.sin_addr,
 378                                                 &client->addr6.sin6_addr))
 379                         goto err;
 380
 381                 client->port = port;
 382         }
 383
 384         client->fd = -1;
 385         client->ops = ops;
 386         client->refs = 1;
 387         client->type = ops->client_type;
 388
 389         __fio_client_add_cmd_option(client, "fio");
 390
 391         flist_add(&client->list, &client_list);
 392         nr_clients++;
 393         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 394         return client;
 395 err:
 396         free(client);
 397         return NULL;
 398 }
 399
 400 int fio_client_add_ini_file(void *cookie, const char *ini_file, bool remote)
 401 {
 402         struct fio_client *client = cookie;
 403         struct client_file *cf;
 404         size_t new_size;
 405         void *new_files;
 406
 407         if (!client)
 408                 return 1;
 409
 410         dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file);
 411
 412         new_size = (client->nr_files + 1) * sizeof(struct client_file);
 413         new_files = realloc(client->files, new_size);
 414         if (!new_files)
 415                 return 1;
 416
 417         client->files = new_files;
 418         cf = &client->files[client->nr_files];
 419         cf->file = strdup(ini_file);
 420         cf->remote = remote;
 421         client->nr_files++;
 422         return 0;
 423 }
 424
 425 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
 426 {
 427         struct fio_client *existing = *cookie;
 428         struct fio_client *client;
 429
 430         if (existing) {
 431                 /*
 432                  * We always add our "exec" name as the option, hence 1
 433                  * means empty.
 434                  */
 435                 if (existing->argc == 1)
 436                         flist_add_tail(&existing->arg_list, &arg_list);
 437                 else {
 438                         while (!flist_empty(&arg_list))
 439                                 flist_del_init(arg_list.next);
 440                 }
 441         }
 442
 443         client = malloc(sizeof(*client));
 444         memset(client, 0, sizeof(*client));
 445
 446         INIT_FLIST_HEAD(&client->list);
 447         INIT_FLIST_HEAD(&client->hash_list);
 448         INIT_FLIST_HEAD(&client->arg_list);
 449         INIT_FLIST_HEAD(&client->eta_list);
 450         INIT_FLIST_HEAD(&client->cmd_list);
 451
 452         if (fio_server_parse_string(hostname, &client->hostname,
 453                                         &client->is_sock, &client->port,
 454                                         &client->addr.sin_addr,
 455                                         &client->addr6.sin6_addr,
 456                                         &client->ipv6))
 457                 return -1;
 458
 459         client->fd = -1;
 460         client->ops = ops;
 461         client->refs = 1;
 462         client->type = ops->client_type;
 463
 464         __fio_client_add_cmd_option(client, "fio");
 465
 466         flist_add(&client->list, &client_list);
 467         nr_clients++;
 468         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 469         *cookie = client;
 470         return 0;
 471 }
 472
 473 static const char *server_name(struct fio_client *client, char *buf,
 474                                size_t bufsize)
 475 {
 476         const char *from;
 477
 478         if (client->ipv6)
 479                 from = inet_ntop(AF_INET6, (struct sockaddr *) &client->addr6.sin6_addr, buf, bufsize);
 480         else if (client->is_sock)
 481                 from = "sock";
 482         else
 483                 from = inet_ntop(AF_INET, (struct sockaddr *) &client->addr.sin_addr, buf, bufsize);
 484
 485         return from;
 486 }
 487
 488 static void probe_client(struct fio_client *client)
 489 {
 490         struct cmd_client_probe_pdu pdu;
 491         const char *sname;
 492         uint64_t tag;
 493         char buf[64];
 494
 495         dprint(FD_NET, "client: send probe\n");
 496
 497 #ifdef CONFIG_ZLIB
 498         pdu.flags = __le64_to_cpu(FIO_PROBE_FLAG_ZLIB);
 499 #else
 500         pdu.flags = 0;
 501 #endif
 502
 503         sname = server_name(client, buf, sizeof(buf));
 504         memset(pdu.server, 0, sizeof(pdu.server));
 505         strncpy((char *) pdu.server, sname, sizeof(pdu.server) - 1);
 506
 507         fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list);
 508 }
 509
 510 static int fio_client_connect_ip(struct fio_client *client)
 511 {
 512         struct sockaddr *addr;
 513         socklen_t socklen;
 514         int fd, domain;
 515
 516         if (client->ipv6) {
 517                 client->addr6.sin6_family = AF_INET6;
 518                 client->addr6.sin6_port = htons(client->port);
 519                 domain = AF_INET6;
 520                 addr = (struct sockaddr *) &client->addr6;
 521                 socklen = sizeof(client->addr6);
 522         } else {
 523                 client->addr.sin_family = AF_INET;
 524                 client->addr.sin_port = htons(client->port);
 525                 domain = AF_INET;
 526                 addr = (struct sockaddr *) &client->addr;
 527                 socklen = sizeof(client->addr);
 528         }
 529
 530         fd = socket(domain, SOCK_STREAM, 0);
 531         if (fd < 0) {
 532                 int ret = -errno;
 533
 534                 log_err("fio: socket: %s\n", strerror(errno));
 535                 return ret;
 536         }
 537
 538         if (connect(fd, addr, socklen) < 0) {
 539                 int ret = -errno;
 540
 541                 log_err("fio: connect: %s\n", strerror(errno));
 542                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 543                                                                 client->port);
 544                 close(fd);
 545                 return ret;
 546         }
 547
 548         return fd;
 549 }
 550
 551 static int fio_client_connect_sock(struct fio_client *client)
 552 {
 553         struct sockaddr_un *addr = &client->addr_un;
 554         socklen_t len;
 555         int fd;
 556
 557         memset(addr, 0, sizeof(*addr));
 558         addr->sun_family = AF_UNIX;
 559         strncpy(addr->sun_path, client->hostname, sizeof(addr->sun_path) - 1);
 560
 561         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 562         if (fd < 0) {
 563                 int ret = -errno;
 564
 565                 log_err("fio: socket: %s\n", strerror(errno));
 566                 return ret;
 567         }
 568
 569         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 570         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 571                 int ret = -errno;
 572
 573                 log_err("fio: connect; %s\n", strerror(errno));
 574                 close(fd);
 575                 return ret;
 576         }
 577
 578         return fd;
 579 }
 580
 581 int fio_client_connect(struct fio_client *client)
 582 {
 583         int fd;
 584
 585         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 586
 587         if (client->is_sock)
 588                 fd = fio_client_connect_sock(client);
 589         else
 590                 fd = fio_client_connect_ip(client);
 591
 592         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 593
 594         if (fd < 0)
 595                 return fd;
 596
 597         client->fd = fd;
 598         fio_client_add_hash(client);
 599         client->state = Client_connected;
 600
 601         probe_client(client);
 602         return 0;
 603 }
 604
 605 int fio_client_terminate(struct fio_client *client)
 606 {
 607         return fio_net_send_quit(client->fd);
 608 }
 609
 610 static void fio_clients_terminate(void)
 611 {
 612         struct flist_head *entry;
 613         struct fio_client *client;
 614
 615         dprint(FD_NET, "client: terminate clients\n");
 616
 617         flist_for_each(entry, &client_list) {
 618                 client = flist_entry(entry, struct fio_client, list);
 619                 fio_client_terminate(client);
 620         }
 621 }
 622
 623 static void sig_int(int sig)
 624 {
 625         dprint(FD_NET, "client: got signal %d\n", sig);
 626         fio_clients_terminate();
 627 }
 628
 629 static void client_signal_handler(void)
 630 {
 631         struct sigaction act;
 632
 633         memset(&act, 0, sizeof(act));
 634         act.sa_handler = sig_int;
 635         act.sa_flags = SA_RESTART;
 636         sigaction(SIGINT, &act, NULL);
 637
 638         memset(&act, 0, sizeof(act));
 639         act.sa_handler = sig_int;
 640         act.sa_flags = SA_RESTART;
 641         sigaction(SIGTERM, &act, NULL);
 642
 643 /* Windows uses SIGBREAK as a quit signal from other applications */
 644 #ifdef WIN32
 645         memset(&act, 0, sizeof(act));
 646         act.sa_handler = sig_int;
 647         act.sa_flags = SA_RESTART;
 648         sigaction(SIGBREAK, &act, NULL);
 649 #endif
 650
 651         memset(&act, 0, sizeof(act));
 652         act.sa_handler = sig_show_status;
 653         act.sa_flags = SA_RESTART;
 654         sigaction(SIGUSR1, &act, NULL);
 655 }
 656
 657 static int send_client_cmd_line(struct fio_client *client)
 658 {
 659         struct cmd_single_line_pdu *cslp;
 660         struct cmd_line_pdu *clp;
 661         unsigned long offset;
 662         unsigned int *lens;
 663         void *pdu;
 664         size_t mem;
 665         int i, ret;
 666
 667         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 668
 669         lens = malloc(client->argc * sizeof(unsigned int));
 670
 671         /*
 672          * Find out how much mem we need
 673          */
 674         for (i = 0, mem = 0; i < client->argc; i++) {
 675                 lens[i] = strlen(client->argv[i]) + 1;
 676                 mem += lens[i];
 677         }
 678
 679         /*
 680          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 681          */
 682         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 683
 684         pdu = malloc(mem);
 685         clp = pdu;
 686         offset = sizeof(*clp);
 687
 688         for (i = 0; i < client->argc; i++) {
 689                 uint16_t arg_len = lens[i];
 690
 691                 cslp = pdu + offset;
 692                 strcpy((char *) cslp->text, client->argv[i]);
 693                 cslp->len = cpu_to_le16(arg_len);
 694                 offset += sizeof(*cslp) + arg_len;
 695         }
 696
 697         free(lens);
 698         clp->lines = cpu_to_le16(client->argc);
 699         clp->client_type = __cpu_to_le16(client->type);
 700         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, NULL, NULL);
 701         free(pdu);
 702         return ret;
 703 }
 704
 705 int fio_clients_connect(void)
 706 {
 707         struct fio_client *client;
 708         struct flist_head *entry, *tmp;
 709         int ret;
 710
 711 #ifdef WIN32
 712         WSADATA wsd;
 713         WSAStartup(MAKEWORD(2, 2), &wsd);
 714 #endif
 715
 716         dprint(FD_NET, "client: connect all\n");
 717
 718         client_signal_handler();
 719
 720         flist_for_each_safe(entry, tmp, &client_list) {
 721                 client = flist_entry(entry, struct fio_client, list);
 722
 723                 ret = fio_client_connect(client);
 724                 if (ret) {
 725                         remove_client(client);
 726                         continue;
 727                 }
 728
 729                 if (client->argc > 1)
 730                         send_client_cmd_line(client);
 731         }
 732
 733         return !nr_clients;
 734 }
 735
 736 int fio_start_client(struct fio_client *client)
 737 {
 738         dprint(FD_NET, "client: start %s\n", client->hostname);
 739         return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
 740 }
 741
 742 int fio_start_all_clients(void)
 743 {
 744         struct fio_client *client;
 745         struct flist_head *entry, *tmp;
 746         int ret;
 747
 748         dprint(FD_NET, "client: start all\n");
 749
 750         fio_client_json_init();
 751
 752         flist_for_each_safe(entry, tmp, &client_list) {
 753                 client = flist_entry(entry, struct fio_client, list);
 754
 755                 ret = fio_start_client(client);
 756                 if (ret) {
 757                         remove_client(client);
 758                         continue;
 759                 }
 760         }
 761
 762         return flist_empty(&client_list);
 763 }
 764
 765 static int __fio_client_send_remote_ini(struct fio_client *client,
 766                                         const char *filename)
 767 {
 768         struct cmd_load_file_pdu *pdu;
 769         size_t p_size;
 770         int ret;
 771
 772         dprint(FD_NET, "send remote ini %s to %s\n", filename, client->hostname);
 773
 774         p_size = sizeof(*pdu) + strlen(filename) + 1;
 775         pdu = malloc(p_size);
 776         memset(pdu, 0, p_size);
 777         pdu->name_len = strlen(filename);
 778         strcpy((char *) pdu->file, filename);
 779         pdu->client_type = cpu_to_le16((uint16_t) client->type);
 780
 781         client->sent_job = true;
 782         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_LOAD_FILE, pdu, p_size,NULL, NULL);
 783         free(pdu);
 784         return ret;
 785 }
 786
 787 /*
 788  * Send file contents to server backend. We could use sendfile(), but to remain
 789  * more portable lets just read/write the darn thing.
 790  */
 791 static int __fio_client_send_local_ini(struct fio_client *client,
 792                                        const char *filename)
 793 {
 794         struct cmd_job_pdu *pdu;
 795         size_t p_size;
 796         struct stat sb;
 797         char *p;
 798         void *buf;
 799         off_t len;
 800         int fd, ret;
 801
 802         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 803
 804         fd = open(filename, O_RDONLY);
 805         if (fd < 0) {
 806                 ret = -errno;
 807                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 808                 return ret;
 809         }
 810
 811         if (fstat(fd, &sb) < 0) {
 812                 ret = -errno;
 813                 log_err("fio: job file stat: %s\n", strerror(errno));
 814                 close(fd);
 815                 return ret;
 816         }
 817
 818         /*
 819          * Add extra space for variable expansion, but doesn't guarantee.
 820          */
 821         sb.st_size += OPT_LEN_MAX;
 822         p_size = sb.st_size + sizeof(*pdu);
 823         pdu = malloc(p_size);
 824         buf = pdu->buf;
 825
 826         len = sb.st_size;
 827         p = buf;
 828         if (read_ini_data(fd, p, len)) {
 829                 log_err("fio: failed reading job file %s\n", filename);
 830                 close(fd);
 831                 free(pdu);
 832                 return 1;
 833         }
 834
 835         pdu->buf_len = __cpu_to_le32(sb.st_size);
 836         pdu->client_type = cpu_to_le32(client->type);
 837
 838         client->sent_job = true;
 839         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, NULL, NULL);
 840         free(pdu);
 841         close(fd);
 842         return ret;
 843 }
 844
 845 int fio_client_send_ini(struct fio_client *client, const char *filename,
 846                         bool remote)
 847 {
 848         int ret;
 849
 850         if (!remote)
 851                 ret = __fio_client_send_local_ini(client, filename);
 852         else
 853                 ret = __fio_client_send_remote_ini(client, filename);
 854
 855         if (!ret)
 856                 client->sent_job = true;
 857
 858         return ret;
 859 }
 860
 861 static int fio_client_send_cf(struct fio_client *client,
 862                               struct client_file *cf)
 863 {
 864         return fio_client_send_ini(client, cf->file, cf->remote);
 865 }
 866
 867 int fio_clients_send_ini(const char *filename)
 868 {
 869         struct fio_client *client;
 870         struct flist_head *entry, *tmp;
 871
 872         flist_for_each_safe(entry, tmp, &client_list) {
 873                 bool failed = false;
 874
 875                 client = flist_entry(entry, struct fio_client, list);
 876
 877                 if (client->nr_files) {
 878                         int i;
 879
 880                         for (i = 0; i < client->nr_files; i++) {
 881                                 struct client_file *cf;
 882
 883                                 cf = &client->files[i];
 884
 885                                 if (fio_client_send_cf(client, cf)) {
 886                                         failed = true;
 887                                         remove_client(client);
 888                                         break;
 889                                 }
 890                         }
 891                 }
 892                 if (client->sent_job || failed)
 893                         continue;
 894                 if (!filename || fio_client_send_ini(client, filename, 0))
 895                         remove_client(client);
 896         }
 897
 898         return !nr_clients;
 899 }
 900
 901 int fio_client_update_options(struct fio_client *client,
 902                               struct thread_options *o, uint64_t *tag)
 903 {
 904         struct cmd_add_job_pdu pdu;
 905
 906         pdu.thread_number = cpu_to_le32(client->thread_number);
 907         pdu.groupid = cpu_to_le32(client->groupid);
 908         convert_thread_options_to_net(&pdu.top, o);
 909
 910         return fio_net_send_cmd(client->fd, FIO_NET_CMD_UPDATE_JOB, &pdu, sizeof(pdu), tag, &client->cmd_list);
 911 }
 912
 913 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 914 {
 915         dst->max_val    = le64_to_cpu(src->max_val);
 916         dst->min_val    = le64_to_cpu(src->min_val);
 917         dst->samples    = le64_to_cpu(src->samples);
 918
 919         /*
 920          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 921          */
 922         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 923         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 924 }
 925
 926 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 927 {
 928         int i, j;
 929
 930         dst->error              = le32_to_cpu(src->error);
 931         dst->thread_number      = le32_to_cpu(src->thread_number);
 932         dst->groupid            = le32_to_cpu(src->groupid);
 933         dst->pid                = le32_to_cpu(src->pid);
 934         dst->members            = le32_to_cpu(src->members);
 935         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
 936
 937         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 938                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 939                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 940                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 941                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 942                 convert_io_stat(&dst->iops_stat[i], &src->iops_stat[i]);
 943         }
 944
 945         dst->usr_time           = le64_to_cpu(src->usr_time);
 946         dst->sys_time           = le64_to_cpu(src->sys_time);
 947         dst->ctx                = le64_to_cpu(src->ctx);
 948         dst->minf               = le64_to_cpu(src->minf);
 949         dst->majf               = le64_to_cpu(src->majf);
 950         dst->clat_percentiles   = le32_to_cpu(src->clat_percentiles);
 951         dst->lat_percentiles    = le32_to_cpu(src->lat_percentiles);
 952         dst->percentile_precision = le64_to_cpu(src->percentile_precision);
 953
 954         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 955                 fio_fp64_t *fps = &src->percentile_list[i];
 956                 fio_fp64_t *fpd = &dst->percentile_list[i];
 957
 958                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 959         }
 960
 961         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 962                 dst->io_u_map[i]        = le64_to_cpu(src->io_u_map[i]);
 963                 dst->io_u_submit[i]     = le64_to_cpu(src->io_u_submit[i]);
 964                 dst->io_u_complete[i]   = le64_to_cpu(src->io_u_complete[i]);
 965         }
 966
 967         for (i = 0; i < FIO_IO_U_LAT_N_NR; i++)
 968                 dst->io_u_lat_n[i]      = le64_to_cpu(src->io_u_lat_n[i]);
 969         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
 970                 dst->io_u_lat_u[i]      = le64_to_cpu(src->io_u_lat_u[i]);
 971         for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
 972                 dst->io_u_lat_m[i]      = le64_to_cpu(src->io_u_lat_m[i]);
 973
 974         for (i = 0; i < DDIR_RWDIR_CNT; i++)
 975                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 976                         dst->io_u_plat[i][j] = le64_to_cpu(src->io_u_plat[i][j]);
 977
 978         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 979                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 980                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 981                 dst->drop_io_u[i]       = le64_to_cpu(src->drop_io_u[i]);
 982         }
 983
 984         dst->total_submit       = le64_to_cpu(src->total_submit);
 985         dst->total_complete     = le64_to_cpu(src->total_complete);
 986
 987         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 988                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 989                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 990         }
 991
 992         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 993         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 994         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 995         dst->first_error        = le32_to_cpu(src->first_error);
 996         dst->kb_base            = le32_to_cpu(src->kb_base);
 997         dst->unit_base          = le32_to_cpu(src->unit_base);
 998
 999         dst->sig_figs           = le32_to_cpu(src->sig_figs);
1000
1001         dst->latency_depth      = le32_to_cpu(src->latency_depth);
1002         dst->latency_target     = le64_to_cpu(src->latency_target);
1003         dst->latency_window     = le64_to_cpu(src->latency_window);
1004         dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
1005
1006         dst->nr_block_infos     = le64_to_cpu(src->nr_block_infos);
1007         for (i = 0; i < dst->nr_block_infos; i++)
1008                 dst->block_infos[i] = le32_to_cpu(src->block_infos[i]);
1009
1010         dst->ss_dur             = le64_to_cpu(src->ss_dur);
1011         dst->ss_state           = le32_to_cpu(src->ss_state);
1012         dst->ss_head            = le32_to_cpu(src->ss_head);
1013         dst->ss_limit.u.f       = fio_uint64_to_double(le64_to_cpu(src->ss_limit.u.i));
1014         dst->ss_slope.u.f       = fio_uint64_to_double(le64_to_cpu(src->ss_slope.u.i));
1015         dst->ss_deviation.u.f   = fio_uint64_to_double(le64_to_cpu(src->ss_deviation.u.i));
1016         dst->ss_criterion.u.f   = fio_uint64_to_double(le64_to_cpu(src->ss_criterion.u.i));
1017
1018         if (dst->ss_state & FIO_SS_DATA) {
1019                 for (i = 0; i < dst->ss_dur; i++ ) {
1020                         dst->ss_iops_data[i] = le64_to_cpu(src->ss_iops_data[i]);
1021                         dst->ss_bw_data[i] = le64_to_cpu(src->ss_bw_data[i]);
1022                 }
1023         }
1024 }
1025
1026 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
1027 {
1028         int i;
1029
1030         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1031                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
1032                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
1033                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
1034                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
1035                 dst->iobytes[i]         = le64_to_cpu(src->iobytes[i]);
1036                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
1037         }
1038
1039         dst->kb_base    = le32_to_cpu(src->kb_base);
1040         dst->unit_base  = le32_to_cpu(src->unit_base);
1041         dst->sig_figs   = le32_to_cpu(src->sig_figs);
1042         dst->groupid    = le32_to_cpu(src->groupid);
1043         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
1044 }
1045
1046 static void json_object_add_client_info(struct json_object *obj,
1047                                         struct fio_client *client)
1048 {
1049         const char *hostname = client->hostname ? client->hostname : "";
1050
1051         json_object_add_value_string(obj, "hostname", hostname);
1052         json_object_add_value_int(obj, "port", client->port);
1053 }
1054
1055 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
1056 {
1057         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1058         struct flist_head *opt_list = NULL;
1059         struct json_object *tsobj;
1060
1061         if (client->opt_lists && p->ts.thread_number <= client->jobs)
1062                 opt_list = &client->opt_lists[p->ts.thread_number - 1];
1063
1064         tsobj = show_thread_status(&p->ts, &p->rs, opt_list, NULL);
1065         client->did_stat = true;
1066         if (tsobj) {
1067                 json_object_add_client_info(tsobj, client);
1068                 json_array_add_value_object(clients_array, tsobj);
1069         }
1070
1071         if (sum_stat_clients <= 1)
1072                 return;
1073
1074         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1);
1075         sum_group_stats(&client_gs, &p->rs);
1076
1077         client_ts.members++;
1078         client_ts.thread_number = p->ts.thread_number;
1079         client_ts.groupid = p->ts.groupid;
1080         client_ts.unified_rw_rep = p->ts.unified_rw_rep;
1081         client_ts.sig_figs = p->ts.sig_figs;
1082
1083         if (++sum_stat_nr == sum_stat_clients) {
1084                 strcpy(client_ts.name, "All clients");
1085                 tsobj = show_thread_status(&client_ts, &client_gs, NULL, NULL);
1086                 if (tsobj) {
1087                         json_object_add_client_info(tsobj, client);
1088                         json_array_add_value_object(clients_array, tsobj);
1089                 }
1090         }
1091 }
1092
1093 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
1094 {
1095         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1096
1097         if (output_format & FIO_OUTPUT_NORMAL)
1098                 show_group_stats(gs, NULL);
1099 }
1100
1101 static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
1102 {
1103         struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
1104         struct print_option *p;
1105
1106         if (!job_opt_object)
1107                 return;
1108
1109         pdu->global = le16_to_cpu(pdu->global);
1110         pdu->truncated = le16_to_cpu(pdu->truncated);
1111         pdu->groupid = le32_to_cpu(pdu->groupid);
1112
1113         p = malloc(sizeof(*p));
1114         p->name = strdup((char *) pdu->name);
1115         if (pdu->value[0] != '\0')
1116                 p->value = strdup((char *) pdu->value);
1117         else
1118                 p->value = NULL;
1119
1120         if (pdu->global) {
1121                 const char *pos = "";
1122
1123                 if (p->value)
1124                         pos = p->value;
1125
1126                 json_object_add_value_string(job_opt_object, p->name, pos);
1127         } else if (client->opt_lists) {
1128                 struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
1129
1130                 flist_add_tail(&p->list, opt_list);
1131         }
1132 }
1133
1134 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
1135 {
1136         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1137         const char *buf = (const char *) pdu->buf;
1138         const char *name;
1139         int fio_unused ret;
1140
1141         name = client->name ? client->name : client->hostname;
1142
1143         if (!client->skip_newline)
1144                 fprintf(f_out, "<%s> ", name);
1145         ret = fwrite(buf, pdu->buf_len, 1, f_out);
1146         fflush(f_out);
1147         client->skip_newline = strchr(buf, '\n') == NULL;
1148 }
1149
1150 static void convert_agg(struct disk_util_agg *agg)
1151 {
1152         int i;
1153
1154         for (i = 0; i < 2; i++) {
1155                 agg->ios[i]     = le64_to_cpu(agg->ios[i]);
1156                 agg->merges[i]  = le64_to_cpu(agg->merges[i]);
1157                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
1158                 agg->ticks[i]   = le64_to_cpu(agg->ticks[i]);
1159         }
1160
1161         agg->io_ticks           = le64_to_cpu(agg->io_ticks);
1162         agg->time_in_queue      = le64_to_cpu(agg->time_in_queue);
1163         agg->slavecount         = le32_to_cpu(agg->slavecount);
1164         agg->max_util.u.f       = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
1165 }
1166
1167 static void convert_dus(struct disk_util_stat *dus)
1168 {
1169         int i;
1170
1171         for (i = 0; i < 2; i++) {
1172                 dus->s.ios[i]           = le64_to_cpu(dus->s.ios[i]);
1173                 dus->s.merges[i]        = le64_to_cpu(dus->s.merges[i]);
1174                 dus->s.sectors[i]       = le64_to_cpu(dus->s.sectors[i]);
1175                 dus->s.ticks[i]         = le64_to_cpu(dus->s.ticks[i]);
1176         }
1177
1178         dus->s.io_ticks         = le64_to_cpu(dus->s.io_ticks);
1179         dus->s.time_in_queue    = le64_to_cpu(dus->s.time_in_queue);
1180         dus->s.msec             = le64_to_cpu(dus->s.msec);
1181 }
1182
1183 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
1184 {
1185         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1186
1187         if (!client->disk_stats_shown) {
1188                 client->disk_stats_shown = true;
1189                 log_info("\nDisk stats (read/write):\n");
1190         }
1191
1192         if (output_format & FIO_OUTPUT_JSON) {
1193                 struct json_object *duobj;
1194                 json_array_add_disk_util(&du->dus, &du->agg, du_array);
1195                 duobj = json_array_last_value_object(du_array);
1196                 json_object_add_client_info(duobj, client);
1197         }
1198         if (output_format & FIO_OUTPUT_TERSE)
1199                 print_disk_util(&du->dus, &du->agg, 1, NULL);
1200         if (output_format & FIO_OUTPUT_NORMAL)
1201                 print_disk_util(&du->dus, &du->agg, 0, NULL);
1202 }
1203
1204 static void convert_jobs_eta(struct jobs_eta *je)
1205 {
1206         int i;
1207
1208         je->nr_running          = le32_to_cpu(je->nr_running);
1209         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
1210         je->nr_pending          = le32_to_cpu(je->nr_pending);
1211         je->nr_setting_up       = le32_to_cpu(je->nr_setting_up);
1212         je->files_open          = le32_to_cpu(je->files_open);
1213
1214         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1215                 je->m_rate[i]   = le64_to_cpu(je->m_rate[i]);
1216                 je->t_rate[i]   = le64_to_cpu(je->t_rate[i]);
1217                 je->m_iops[i]   = le32_to_cpu(je->m_iops[i]);
1218                 je->t_iops[i]   = le32_to_cpu(je->t_iops[i]);
1219                 je->rate[i]     = le64_to_cpu(je->rate[i]);
1220                 je->iops[i]     = le32_to_cpu(je->iops[i]);
1221         }
1222
1223         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
1224         je->eta_sec             = le64_to_cpu(je->eta_sec);
1225         je->nr_threads          = le32_to_cpu(je->nr_threads);
1226         je->is_pow2             = le32_to_cpu(je->is_pow2);
1227         je->unit_base           = le32_to_cpu(je->unit_base);
1228         je->sig_figs            = le32_to_cpu(je->sig_figs);
1229 }
1230
1231 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
1232 {
1233         int i;
1234
1235         dst->nr_running         += je->nr_running;
1236         dst->nr_ramp            += je->nr_ramp;
1237         dst->nr_pending         += je->nr_pending;
1238         dst->nr_setting_up      += je->nr_setting_up;
1239         dst->files_open         += je->files_open;
1240
1241         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1242                 dst->m_rate[i]  += je->m_rate[i];
1243                 dst->t_rate[i]  += je->t_rate[i];
1244                 dst->m_iops[i]  += je->m_iops[i];
1245                 dst->t_iops[i]  += je->t_iops[i];
1246                 dst->rate[i]    += je->rate[i];
1247                 dst->iops[i]    += je->iops[i];
1248         }
1249
1250         dst->elapsed_sec        += je->elapsed_sec;
1251
1252         if (je->eta_sec > dst->eta_sec)
1253                 dst->eta_sec = je->eta_sec;
1254
1255         dst->nr_threads         += je->nr_threads;
1256
1257         /*
1258          * This wont be correct for multiple strings, but at least it
1259          * works for the basic cases.
1260          */
1261         strcpy((char *) dst->run_str, (char *) je->run_str);
1262 }
1263
1264 static bool remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
1265 {
1266         struct fio_net_cmd_reply *reply = NULL;
1267         struct flist_head *entry;
1268
1269         flist_for_each(entry, &client->cmd_list) {
1270                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1271
1272                 if (cmd->tag == (uintptr_t) reply)
1273                         break;
1274
1275                 reply = NULL;
1276         }
1277
1278         if (!reply) {
1279                 log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag);
1280                 return false;
1281         }
1282
1283         flist_del(&reply->list);
1284         cmd->tag = reply->saved_tag;
1285         free(reply);
1286         return true;
1287 }
1288
1289 int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
1290 {
1291         do {
1292                 struct fio_net_cmd_reply *reply = NULL;
1293                 struct flist_head *entry;
1294
1295                 flist_for_each(entry, &client->cmd_list) {
1296                         reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1297
1298                         if (tag == (uintptr_t) reply)
1299                                 break;
1300
1301                         reply = NULL;
1302                 }
1303
1304                 if (!reply)
1305                         break;
1306
1307                 usleep(1000);
1308         } while (1);
1309
1310         return 0;
1311 }
1312
1313 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
1314 {
1315         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1316         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
1317
1318         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
1319
1320         assert(client->eta_in_flight == eta);
1321
1322         client->eta_in_flight = NULL;
1323         flist_del_init(&client->eta_list);
1324         client->eta_timeouts = 0;
1325
1326         if (client->ops->jobs_eta)
1327                 client->ops->jobs_eta(client, je);
1328
1329         fio_client_sum_jobs_eta(&eta->eta, je);
1330         fio_client_dec_jobs_eta(eta, client->ops->eta);
1331 }
1332
1333 static void client_flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
1334                                       uint64_t sample_size)
1335 {
1336         struct io_sample *s;
1337         int log_offset;
1338         uint64_t i, j, nr_samples;
1339         struct io_u_plat_entry *entry;
1340         uint64_t *io_u_plat;
1341
1342         int stride = 1 << hist_coarseness;
1343
1344         if (!sample_size)
1345                 return;
1346
1347         s = __get_sample(samples, 0, 0);
1348         log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
1349
1350         nr_samples = sample_size / __log_entry_sz(log_offset);
1351
1352         for (i = 0; i < nr_samples; i++) {
1353
1354                 s = (struct io_sample *)((char *)__get_sample(samples, log_offset, i) +
1355                         i * sizeof(struct io_u_plat_entry));
1356
1357                 entry = s->data.plat_entry;
1358                 io_u_plat = entry->io_u_plat;
1359
1360                 fprintf(f, "%lu, %u, %llu, ", (unsigned long) s->time,
1361                                                 io_sample_ddir(s), (unsigned long long) s->bs);
1362                 for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) {
1363                         fprintf(f, "%llu, ", (unsigned long long)hist_sum(j, stride, io_u_plat, NULL));
1364                 }
1365                 fprintf(f, "%llu\n", (unsigned long long)
1366                         hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat, NULL));
1367
1368         }
1369 }
1370
1371 static int fio_client_handle_iolog(struct fio_client *client,
1372                                    struct fio_net_cmd *cmd)
1373 {
1374         struct cmd_iolog_pdu *pdu = NULL;
1375         bool store_direct;
1376         char *log_pathname = NULL;
1377         int ret = 0;
1378
1379         pdu = convert_iolog(cmd, &store_direct);
1380         if (!pdu) {
1381                 log_err("fio: failed converting IO log\n");
1382                 ret = 1;
1383                 goto out;
1384         }
1385
1386         /* allocate buffer big enough for next sprintf() call */
1387         log_pathname = malloc(10 + strlen((char *)pdu->name) +
1388                         strlen(client->hostname));
1389         if (!log_pathname) {
1390                 log_err("fio: memory allocation of unique pathname failed\n");
1391                 ret = -1;
1392                 goto out;
1393         }
1394         /* generate a unique pathname for the log file using hostname */
1395         sprintf(log_pathname, "%s.%s", pdu->name, client->hostname);
1396
1397         if (store_direct) {
1398                 ssize_t wrote;
1399                 size_t sz;
1400                 int fd;
1401
1402                 fd = open((const char *) log_pathname,
1403                                 O_WRONLY | O_CREAT | O_TRUNC, 0644);
1404                 if (fd < 0) {
1405                         log_err("fio: open log %s: %s\n",
1406                                 log_pathname, strerror(errno));
1407                         ret = 1;
1408                         goto out;
1409                 }
1410
1411                 sz = cmd->pdu_len - sizeof(*pdu);
1412                 wrote = write(fd, pdu->samples, sz);
1413                 close(fd);
1414
1415                 if (wrote != sz) {
1416                         log_err("fio: short write on compressed log\n");
1417                         ret = 1;
1418                         goto out;
1419                 }
1420
1421                 ret = 0;
1422         } else {
1423                 FILE *f;
1424                 f = fopen((const char *) log_pathname, "w");
1425                 if (!f) {
1426                         log_err("fio: fopen log %s : %s\n",
1427                                 log_pathname, strerror(errno));
1428                         ret = 1;
1429                         goto out;
1430                 }
1431
1432                 if (pdu->log_type == IO_LOG_TYPE_HIST) {
1433                         client_flush_hist_samples(f, pdu->log_hist_coarseness, pdu->samples,
1434                                            pdu->nr_samples * sizeof(struct io_sample));
1435                 } else {
1436                         flush_samples(f, pdu->samples,
1437                                         pdu->nr_samples * sizeof(struct io_sample));
1438                 }
1439                 fclose(f);
1440                 ret = 0;
1441         }
1442
1443 out:
1444         if (pdu && pdu != (void *) cmd->payload)
1445                 free(pdu);
1446
1447         if (log_pathname)
1448                 free(log_pathname);
1449
1450         return ret;
1451 }
1452
1453 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
1454 {
1455         struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
1456         const char *os, *arch;
1457         char bit[16];
1458
1459         os = fio_get_os_string(probe->os);
1460         if (!os)
1461                 os = "unknown";
1462
1463         arch = fio_get_arch_string(probe->arch);
1464         if (!arch)
1465                 os = "unknown";
1466
1467         sprintf(bit, "%d-bit", probe->bpp * 8);
1468         probe->flags = le64_to_cpu(probe->flags);
1469
1470         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n",
1471                 probe->hostname, probe->bigendian, bit, os, arch,
1472                 probe->fio_version, (unsigned long) probe->flags);
1473
1474         if (!client->name)
1475                 client->name = strdup((char *) probe->hostname);
1476 }
1477
1478 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
1479 {
1480         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1481
1482         client->state = Client_started;
1483         client->jobs = le32_to_cpu(pdu->jobs);
1484         client->nr_stat = le32_to_cpu(pdu->stat_outputs);
1485
1486         if (client->jobs) {
1487                 int i;
1488
1489                 if (client->opt_lists)
1490                         free(client->opt_lists);
1491
1492                 client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
1493                 for (i = 0; i < client->jobs; i++)
1494                         INIT_FLIST_HEAD(&client->opt_lists[i]);
1495         }
1496
1497         sum_stat_clients += client->nr_stat;
1498 }
1499
1500 static void handle_stop(struct fio_client *client)
1501 {
1502         if (client->error)
1503                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
1504 }
1505
1506 static void convert_stop(struct fio_net_cmd *cmd)
1507 {
1508         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1509
1510         pdu->error = le32_to_cpu(pdu->error);
1511 }
1512
1513 static void convert_text(struct fio_net_cmd *cmd)
1514 {
1515         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1516
1517         pdu->level      = le32_to_cpu(pdu->level);
1518         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
1519         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
1520         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
1521 }
1522
1523 static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
1524                                               struct cmd_iolog_pdu *pdu)
1525 {
1526 #ifdef CONFIG_ZLIB
1527         struct cmd_iolog_pdu *ret;
1528         z_stream stream;
1529         uint32_t nr_samples;
1530         size_t total;
1531         char *p;
1532
1533         stream.zalloc = Z_NULL;
1534         stream.zfree = Z_NULL;
1535         stream.opaque = Z_NULL;
1536         stream.avail_in = 0;
1537         stream.next_in = Z_NULL;
1538
1539         if (inflateInit(&stream) != Z_OK)
1540                 return NULL;
1541
1542         /*
1543          * Get header first, it's not compressed
1544          */
1545         nr_samples = le64_to_cpu(pdu->nr_samples);
1546
1547         if (pdu->log_type == IO_LOG_TYPE_HIST)
1548                 total = nr_samples * (__log_entry_sz(le32_to_cpu(pdu->log_offset)) +
1549                                         sizeof(struct io_u_plat_entry));
1550         else
1551                 total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
1552         ret = malloc(total + sizeof(*pdu));
1553         ret->nr_samples = nr_samples;
1554
1555         memcpy(ret, pdu, sizeof(*pdu));
1556
1557         p = (char *) ret + sizeof(*pdu);
1558
1559         stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1560         stream.next_in = (void *)((char *) pdu + sizeof(*pdu));
1561         while (stream.avail_in) {
1562                 unsigned int this_chunk = 65536;
1563                 unsigned int this_len;
1564                 int err;
1565
1566                 if (this_chunk > total)
1567                         this_chunk = total;
1568
1569                 stream.avail_out = this_chunk;
1570                 stream.next_out = (void *)p;
1571                 err = inflate(&stream, Z_NO_FLUSH);
1572                 /* may be Z_OK, or Z_STREAM_END */
1573                 if (err < 0) {
1574                         log_err("fio: inflate error %d\n", err);
1575                         free(ret);
1576                         ret = NULL;
1577                         goto err;
1578                 }
1579
1580                 this_len = this_chunk - stream.avail_out;
1581                 p += this_len;
1582                 total -= this_len;
1583         }
1584
1585 err:
1586         inflateEnd(&stream);
1587         return ret;
1588 #else
1589         return NULL;
1590 #endif
1591 }
1592
1593 /*
1594  * This has been compressed on the server side, since it can be big.
1595  * Uncompress here.
1596  */
1597 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd,
1598                                            bool *store_direct)
1599 {
1600         struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1601         struct cmd_iolog_pdu *ret;
1602         uint64_t i;
1603         int compressed;
1604         void *samples;
1605
1606         *store_direct = false;
1607
1608         /*
1609          * Convert if compressed and we support it. If it's not
1610          * compressed, we need not do anything.
1611          */
1612         compressed = le32_to_cpu(pdu->compressed);
1613         if (compressed == XMIT_COMPRESSED) {
1614 #ifndef CONFIG_ZLIB
1615                 log_err("fio: server sent compressed data by mistake\n");
1616                 return NULL;
1617 #endif
1618                 ret = convert_iolog_gz(cmd, pdu);
1619                 if (!ret) {
1620                         log_err("fio: failed decompressing log\n");
1621                         return NULL;
1622                 }
1623         } else if (compressed == STORE_COMPRESSED) {
1624                 *store_direct = true;
1625                 ret = pdu;
1626         } else
1627                 ret = pdu;
1628
1629         ret->nr_samples         = le64_to_cpu(ret->nr_samples);
1630         ret->thread_number      = le32_to_cpu(ret->thread_number);
1631         ret->log_type           = le32_to_cpu(ret->log_type);
1632         ret->compressed         = le32_to_cpu(ret->compressed);
1633         ret->log_offset         = le32_to_cpu(ret->log_offset);
1634         ret->log_hist_coarseness = le32_to_cpu(ret->log_hist_coarseness);
1635
1636         if (*store_direct)
1637                 return ret;
1638
1639         samples = &ret->samples[0];
1640         for (i = 0; i < ret->nr_samples; i++) {
1641                 struct io_sample *s;
1642
1643                 s = __get_sample(samples, ret->log_offset, i);
1644                 if (ret->log_type == IO_LOG_TYPE_HIST)
1645                         s = (struct io_sample *)((char *)s + sizeof(struct io_u_plat_entry) * i);
1646
1647                 s->time         = le64_to_cpu(s->time);
1648                 s->data.val     = le64_to_cpu(s->data.val);
1649                 s->__ddir       = le32_to_cpu(s->__ddir);
1650                 s->bs           = le64_to_cpu(s->bs);
1651
1652                 if (ret->log_offset) {
1653                         struct io_sample_offset *so = (void *) s;
1654
1655                         so->offset = le64_to_cpu(so->offset);
1656                 }
1657
1658                 if (ret->log_type == IO_LOG_TYPE_HIST) {
1659                         s->data.plat_entry = (struct io_u_plat_entry *)(((char *)s) + sizeof(*s));
1660                         s->data.plat_entry->list.next = NULL;
1661                         s->data.plat_entry->list.prev = NULL;
1662                 }
1663         }
1664
1665         return ret;
1666 }
1667
1668 static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep,
1669                            size_t size, uint64_t tag)
1670 {
1671         rep->error = cpu_to_le32(rep->error);
1672         fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL);
1673 }
1674
1675 static int fio_send_file(struct fio_client *client, struct cmd_sendfile *pdu,
1676                          uint64_t tag)
1677 {
1678         struct cmd_sendfile_reply *rep;
1679         struct stat sb;
1680         size_t size;
1681         int fd;
1682
1683         size = sizeof(*rep);
1684         rep = malloc(size);
1685
1686         if (stat((char *)pdu->path, &sb) < 0) {
1687 fail:
1688                 rep->error = errno;
1689                 sendfile_reply(client->fd, rep, size, tag);
1690                 free(rep);
1691                 return 1;
1692         }
1693
1694         size += sb.st_size;
1695         rep = realloc(rep, size);
1696         rep->size = cpu_to_le32((uint32_t) sb.st_size);
1697
1698         fd = open((char *)pdu->path, O_RDONLY);
1699         if (fd == -1 )
1700                 goto fail;
1701
1702         rep->error = read_data(fd, &rep->data, sb.st_size);
1703         sendfile_reply(client->fd, rep, size, tag);
1704         free(rep);
1705         close(fd);
1706         return 0;
1707 }
1708
1709 int fio_handle_client(struct fio_client *client)
1710 {
1711         struct client_ops *ops = client->ops;
1712         struct fio_net_cmd *cmd;
1713         int size;
1714
1715         dprint(FD_NET, "client: handle %s\n", client->hostname);
1716
1717         cmd = fio_net_recv_cmd(client->fd, true);
1718         if (!cmd)
1719                 return 0;
1720
1721         dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1722                 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1723
1724         client->last_cmd = cmd->opcode;
1725
1726         switch (cmd->opcode) {
1727         case FIO_NET_CMD_QUIT:
1728                 if (ops->quit)
1729                         ops->quit(client, cmd);
1730                 remove_client(client);
1731                 break;
1732         case FIO_NET_CMD_TEXT:
1733                 convert_text(cmd);
1734                 ops->text(client, cmd);
1735                 break;
1736         case FIO_NET_CMD_DU: {
1737                 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1738
1739                 convert_dus(&du->dus);
1740                 convert_agg(&du->agg);
1741
1742                 ops->disk_util(client, cmd);
1743                 break;
1744                 }
1745         case FIO_NET_CMD_TS: {
1746                 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1747
1748                 dprint(FD_NET, "client: ts->ss_state = %u\n", (unsigned int) le32_to_cpu(p->ts.ss_state));
1749                 if (le32_to_cpu(p->ts.ss_state) & FIO_SS_DATA) {
1750                         dprint(FD_NET, "client: received steadystate ring buffers\n");
1751
1752                         size = le64_to_cpu(p->ts.ss_dur);
1753                         p->ts.ss_iops_data = (uint64_t *) ((struct cmd_ts_pdu *)cmd->payload + 1);
1754                         p->ts.ss_bw_data = p->ts.ss_iops_data + size;
1755                 }
1756
1757                 convert_ts(&p->ts, &p->ts);
1758                 convert_gs(&p->rs, &p->rs);
1759
1760                 ops->thread_status(client, cmd);
1761                 break;
1762                 }
1763         case FIO_NET_CMD_GS: {
1764                 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1765
1766                 convert_gs(gs, gs);
1767
1768                 ops->group_stats(client, cmd);
1769                 break;
1770                 }
1771         case FIO_NET_CMD_ETA: {
1772                 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1773
1774                 if (!remove_reply_cmd(client, cmd))
1775                         break;
1776                 convert_jobs_eta(je);
1777                 handle_eta(client, cmd);
1778                 break;
1779                 }
1780         case FIO_NET_CMD_PROBE:
1781                 remove_reply_cmd(client, cmd);
1782                 ops->probe(client, cmd);
1783                 break;
1784         case FIO_NET_CMD_SERVER_START:
1785                 client->state = Client_running;
1786                 if (ops->job_start)
1787                         ops->job_start(client, cmd);
1788                 break;
1789         case FIO_NET_CMD_START: {
1790                 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1791
1792                 pdu->jobs = le32_to_cpu(pdu->jobs);
1793                 ops->start(client, cmd);
1794                 break;
1795                 }
1796         case FIO_NET_CMD_STOP: {
1797                 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1798
1799                 convert_stop(cmd);
1800                 client->state = Client_stopped;
1801                 client->error = le32_to_cpu(pdu->error);
1802                 client->signal = le32_to_cpu(pdu->signal);
1803                 ops->stop(client);
1804                 break;
1805                 }
1806         case FIO_NET_CMD_ADD_JOB: {
1807                 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1808
1809                 client->thread_number = le32_to_cpu(pdu->thread_number);
1810                 client->groupid = le32_to_cpu(pdu->groupid);
1811
1812                 if (ops->add_job)
1813                         ops->add_job(client, cmd);
1814                 break;
1815                 }
1816         case FIO_NET_CMD_IOLOG:
1817                 fio_client_handle_iolog(client, cmd);
1818                 break;
1819         case FIO_NET_CMD_UPDATE_JOB:
1820                 ops->update_job(client, cmd);
1821                 remove_reply_cmd(client, cmd);
1822                 break;
1823         case FIO_NET_CMD_VTRIGGER: {
1824                 struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
1825                 char buf[128];
1826                 int off = 0;
1827
1828                 if (aux_path) {
1829                         strcpy(buf, aux_path);
1830                         off = strlen(buf);
1831                 }
1832
1833                 __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
1834                 exec_trigger(trigger_cmd);
1835                 break;
1836                 }
1837         case FIO_NET_CMD_SENDFILE: {
1838                 struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload;
1839                 fio_send_file(client, pdu, cmd->tag);
1840                 break;
1841                 }
1842         case FIO_NET_CMD_JOB_OPT: {
1843                 handle_job_opt(client, cmd);
1844                 break;
1845         }
1846         default:
1847                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1848                 break;
1849         }
1850
1851         free(cmd);
1852         return 1;
1853 }
1854
1855 int fio_clients_send_trigger(const char *cmd)
1856 {
1857         struct flist_head *entry;
1858         struct fio_client *client;
1859         size_t slen;
1860
1861         dprint(FD_NET, "client: send vtrigger: %s\n", cmd);
1862
1863         if (!cmd)
1864                 slen = 0;
1865         else
1866                 slen = strlen(cmd);
1867
1868         flist_for_each(entry, &client_list) {
1869                 struct cmd_vtrigger_pdu *pdu;
1870
1871                 client = flist_entry(entry, struct fio_client, list);
1872
1873                 pdu = malloc(sizeof(*pdu) + slen);
1874                 pdu->len = cpu_to_le16((uint16_t) slen);
1875                 if (slen)
1876                         memcpy(pdu->cmd, cmd, slen);
1877                 fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu,
1878                                         sizeof(*pdu) + slen, NULL, NULL);
1879                 free(pdu);
1880         }
1881
1882         return 0;
1883 }
1884
1885 static void request_client_etas(struct client_ops *ops)
1886 {
1887         struct fio_client *client;
1888         struct flist_head *entry;
1889         struct client_eta *eta;
1890         int skipped = 0;
1891
1892         if (eta_print == FIO_ETA_NEVER)
1893                 return;
1894
1895         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1896
1897         eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
1898         eta->pending = nr_clients;
1899
1900         flist_for_each(entry, &client_list) {
1901                 client = flist_entry(entry, struct fio_client, list);
1902
1903                 if (!flist_empty(&client->eta_list)) {
1904                         skipped++;
1905                         continue;
1906                 }
1907                 if (client->state != Client_running)
1908                         continue;
1909
1910                 assert(!client->eta_in_flight);
1911                 flist_add_tail(&client->eta_list, &eta_list);
1912                 client->eta_in_flight = eta;
1913                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1914                                         (uintptr_t) eta, &client->cmd_list);
1915         }
1916
1917         while (skipped--) {
1918                 if (!fio_client_dec_jobs_eta(eta, ops->eta))
1919                         break;
1920         }
1921
1922         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1923 }
1924
1925 /*
1926  * A single SEND_ETA timeout isn't fatal. Attempt to recover.
1927  */
1928 static int handle_cmd_timeout(struct fio_client *client,
1929                               struct fio_net_cmd_reply *reply)
1930 {
1931         uint16_t reply_opcode = reply->opcode;
1932
1933         flist_del(&reply->list);
1934         free(reply);
1935
1936         if (reply_opcode != FIO_NET_CMD_SEND_ETA)
1937                 return 1;
1938
1939         log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
1940
1941         flist_del_init(&client->eta_list);
1942         if (client->eta_in_flight) {
1943                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
1944                 client->eta_in_flight = NULL;
1945         }
1946
1947         /*
1948          * If we fail 5 in a row, give up...
1949          */
1950         if (client->eta_timeouts++ > 5)
1951                 return 1;
1952
1953         return 0;
1954 }
1955
1956 static int client_check_cmd_timeout(struct fio_client *client,
1957                                     struct timespec *now)
1958 {
1959         struct fio_net_cmd_reply *reply;
1960         struct flist_head *entry, *tmp;
1961         int ret = 0;
1962
1963         flist_for_each_safe(entry, tmp, &client->cmd_list) {
1964                 unsigned int op;
1965
1966                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1967
1968                 if (mtime_since(&reply->ts, now) < FIO_NET_CLIENT_TIMEOUT)
1969                         continue;
1970
1971                 op = reply->opcode;
1972                 if (!handle_cmd_timeout(client, reply))
1973                         continue;
1974
1975                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1976                                                 fio_server_op(op));
1977                 ret = 1;
1978         }
1979
1980         return flist_empty(&client->cmd_list) && ret;
1981 }
1982
1983 static int fio_check_clients_timed_out(void)
1984 {
1985         struct fio_client *client;
1986         struct flist_head *entry, *tmp;
1987         struct timespec ts;
1988         int ret = 0;
1989
1990         fio_gettime(&ts, NULL);
1991
1992         flist_for_each_safe(entry, tmp, &client_list) {
1993                 client = flist_entry(entry, struct fio_client, list);
1994
1995                 if (flist_empty(&client->cmd_list))
1996                         continue;
1997
1998                 if (!client_check_cmd_timeout(client, &ts))
1999                         continue;
2000
2001                 if (client->ops->timed_out)
2002                         client->ops->timed_out(client);
2003                 else
2004                         log_err("fio: client %s timed out\n", client->hostname);
2005
2006                 if (client->last_cmd != FIO_NET_CMD_VTRIGGER)
2007                         client->error = ETIMEDOUT;
2008                 else
2009                         log_info("fio: ignoring timeout due to vtrigger\n");
2010                 remove_client(client);
2011                 ret = 1;
2012         }
2013
2014         return ret;
2015 }
2016
2017 int fio_handle_clients(struct client_ops *ops)
2018 {
2019         struct pollfd *pfds;
2020         int i, ret = 0, retval = 0;
2021
2022         fio_gettime(&eta_ts, NULL);
2023
2024         pfds = malloc(nr_clients * sizeof(struct pollfd));
2025
2026         init_thread_stat(&client_ts);
2027         init_group_run_stat(&client_gs);
2028
2029         while (!exit_backend && nr_clients) {
2030                 struct flist_head *entry, *tmp;
2031                 struct fio_client *client;
2032
2033                 i = 0;
2034                 flist_for_each_safe(entry, tmp, &client_list) {
2035                         client = flist_entry(entry, struct fio_client, list);
2036
2037                         if (!client->sent_job && !client->ops->stay_connected &&
2038                             flist_empty(&client->cmd_list)) {
2039                                 remove_client(client);
2040                                 continue;
2041                         }
2042
2043                         pfds[i].fd = client->fd;
2044                         pfds[i].events = POLLIN;
2045                         i++;
2046                 }
2047
2048                 if (!nr_clients)
2049                         break;
2050
2051                 assert(i == nr_clients);
2052
2053                 do {
2054                         struct timespec ts;
2055                         int timeout;
2056
2057                         fio_gettime(&ts, NULL);
2058                         if (eta_time_within_slack(mtime_since(&eta_ts, &ts))) {
2059                                 request_client_etas(ops);
2060                                 memcpy(&eta_ts, &ts, sizeof(ts));
2061
2062                                 if (fio_check_clients_timed_out())
2063                                         break;
2064                         }
2065
2066                         check_trigger_file();
2067
2068                         timeout = min(100u, ops->eta_msec);
2069
2070                         ret = poll(pfds, nr_clients, timeout);
2071                         if (ret < 0) {
2072                                 if (errno == EINTR)
2073                                         continue;
2074                                 log_err("fio: poll clients: %s\n", strerror(errno));
2075                                 break;
2076                         } else if (!ret)
2077                                 continue;
2078                 } while (ret <= 0);
2079
2080                 for (i = 0; i < nr_clients; i++) {
2081                         if (!(pfds[i].revents & POLLIN))
2082                                 continue;
2083
2084                         client = find_client_by_fd(pfds[i].fd);
2085                         if (!client) {
2086                                 log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd);
2087                                 continue;
2088                         }
2089                         if (!fio_handle_client(client)) {
2090                                 log_info("client: host=%s disconnected\n",
2091                                                 client->hostname);
2092                                 remove_client(client);
2093                                 retval = 1;
2094                         } else if (client->error)
2095                                 retval = 1;
2096                         fio_put_client(client);
2097                 }
2098         }
2099
2100         fio_client_json_fini();
2101
2102         free(pfds);
2103         return retval || error_clients;
2104 }