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