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