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