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