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 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         dst->ss_dur             = le64_to_cpu(src->ss_dur);
 951         dst->ss_state           = le32_to_cpu(src->ss_state);
 952         dst->ss_head            = le32_to_cpu(src->ss_head);
 953         dst->ss_limit.u.f       = fio_uint64_to_double(le64_to_cpu(src->ss_limit.u.i));
 954         dst->ss_slope.u.f       = fio_uint64_to_double(le64_to_cpu(src->ss_slope.u.i));
 955         dst->ss_deviation.u.f   = fio_uint64_to_double(le64_to_cpu(src->ss_deviation.u.i));
 956         dst->ss_criterion.u.f   = fio_uint64_to_double(le64_to_cpu(src->ss_criterion.u.i));
 957
 958         if (dst->ss_state & __FIO_SS_DATA) {
 959                 for (i = 0; i < dst->ss_dur; i++ ) {
 960                         dst->ss_iops_data[i] = le64_to_cpu(src->ss_iops_data[i]);
 961                         dst->ss_bw_data[i] = le64_to_cpu(src->ss_bw_data[i]);
 962                 }
 963         }
 964 }
 965
 966 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 967 {
 968         int i;
 969
 970         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 971                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 972                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 973                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 974                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 975                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 976                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 977         }
 978
 979         dst->kb_base    = le32_to_cpu(src->kb_base);
 980         dst->unit_base  = le32_to_cpu(src->unit_base);
 981         dst->groupid    = le32_to_cpu(src->groupid);
 982         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
 983 }
 984
 985 static void json_object_add_client_info(struct json_object *obj,
 986                                         struct fio_client *client)
 987 {
 988         const char *hostname = client->hostname ? client->hostname : "";
 989
 990         json_object_add_value_string(obj, "hostname", hostname);
 991         json_object_add_value_int(obj, "port", client->port);
 992 }
 993
 994 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
 995 {
 996         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 997         struct flist_head *opt_list = NULL;
 998         struct json_object *tsobj;
 999
1000         if (client->opt_lists && p->ts.thread_number <= client->jobs)
1001                 opt_list = &client->opt_lists[p->ts.thread_number - 1];
1002
1003         tsobj = show_thread_status(&p->ts, &p->rs, opt_list, NULL);
1004         client->did_stat = 1;
1005         if (tsobj) {
1006                 json_object_add_client_info(tsobj, client);
1007                 json_array_add_value_object(clients_array, tsobj);
1008         }
1009
1010         if (sum_stat_clients <= 1)
1011                 return;
1012
1013         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1);
1014         sum_group_stats(&client_gs, &p->rs);
1015
1016         client_ts.members++;
1017         client_ts.thread_number = p->ts.thread_number;
1018         client_ts.groupid = p->ts.groupid;
1019         client_ts.unified_rw_rep = p->ts.unified_rw_rep;
1020
1021         if (++sum_stat_nr == sum_stat_clients) {
1022                 strcpy(client_ts.name, "All clients");
1023                 tsobj = show_thread_status(&client_ts, &client_gs, NULL, NULL);
1024                 if (tsobj) {
1025                         json_object_add_client_info(tsobj, client);
1026                         json_array_add_value_object(clients_array, tsobj);
1027                 }
1028         }
1029 }
1030
1031 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
1032 {
1033         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1034
1035         if (output_format & FIO_OUTPUT_NORMAL)
1036                 show_group_stats(gs, NULL);
1037 }
1038
1039 static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
1040 {
1041         struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
1042         struct print_option *p;
1043
1044         if (!job_opt_object)
1045                 return;
1046
1047         pdu->global = le16_to_cpu(pdu->global);
1048         pdu->truncated = le16_to_cpu(pdu->truncated);
1049         pdu->groupid = le32_to_cpu(pdu->groupid);
1050
1051         p = malloc(sizeof(*p));
1052         p->name = strdup((char *) pdu->name);
1053         if (pdu->value[0] != '\0')
1054                 p->value = strdup((char *) pdu->value);
1055         else
1056                 p->value = NULL;
1057
1058         if (pdu->global) {
1059                 const char *pos = "";
1060
1061                 if (p->value)
1062                         pos = p->value;
1063
1064                 json_object_add_value_string(job_opt_object, p->name, pos);
1065         } else if (client->opt_lists) {
1066                 struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
1067
1068                 flist_add_tail(&p->list, opt_list);
1069         }
1070 }
1071
1072 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
1073 {
1074         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1075         const char *buf = (const char *) pdu->buf;
1076         const char *name;
1077         int fio_unused ret;
1078
1079         name = client->name ? client->name : client->hostname;
1080
1081         if (!client->skip_newline)
1082                 fprintf(f_out, "<%s> ", name);
1083         ret = fwrite(buf, pdu->buf_len, 1, f_out);
1084         fflush(f_out);
1085         client->skip_newline = strchr(buf, '\n') == NULL;
1086 }
1087
1088 static void convert_agg(struct disk_util_agg *agg)
1089 {
1090         int i;
1091
1092         for (i = 0; i < 2; i++) {
1093                 agg->ios[i]     = le64_to_cpu(agg->ios[i]);
1094                 agg->merges[i]  = le64_to_cpu(agg->merges[i]);
1095                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
1096                 agg->ticks[i]   = le64_to_cpu(agg->ticks[i]);
1097         }
1098
1099         agg->io_ticks           = le64_to_cpu(agg->io_ticks);
1100         agg->time_in_queue      = le64_to_cpu(agg->time_in_queue);
1101         agg->slavecount         = le32_to_cpu(agg->slavecount);
1102         agg->max_util.u.f       = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
1103 }
1104
1105 static void convert_dus(struct disk_util_stat *dus)
1106 {
1107         int i;
1108
1109         for (i = 0; i < 2; i++) {
1110                 dus->s.ios[i]           = le64_to_cpu(dus->s.ios[i]);
1111                 dus->s.merges[i]        = le64_to_cpu(dus->s.merges[i]);
1112                 dus->s.sectors[i]       = le64_to_cpu(dus->s.sectors[i]);
1113                 dus->s.ticks[i]         = le64_to_cpu(dus->s.ticks[i]);
1114         }
1115
1116         dus->s.io_ticks         = le64_to_cpu(dus->s.io_ticks);
1117         dus->s.time_in_queue    = le64_to_cpu(dus->s.time_in_queue);
1118         dus->s.msec             = le64_to_cpu(dus->s.msec);
1119 }
1120
1121 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
1122 {
1123         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1124
1125         if (!client->disk_stats_shown) {
1126                 client->disk_stats_shown = 1;
1127                 log_info("\nDisk stats (read/write):\n");
1128         }
1129
1130         if (output_format & FIO_OUTPUT_JSON) {
1131                 struct json_object *duobj;
1132                 json_array_add_disk_util(&du->dus, &du->agg, du_array);
1133                 duobj = json_array_last_value_object(du_array);
1134                 json_object_add_client_info(duobj, client);
1135         }
1136         if (output_format & FIO_OUTPUT_TERSE)
1137                 print_disk_util(&du->dus, &du->agg, 1, NULL);
1138         if (output_format & FIO_OUTPUT_NORMAL)
1139                 print_disk_util(&du->dus, &du->agg, 0, NULL);
1140 }
1141
1142 static void convert_jobs_eta(struct jobs_eta *je)
1143 {
1144         int i;
1145
1146         je->nr_running          = le32_to_cpu(je->nr_running);
1147         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
1148         je->nr_pending          = le32_to_cpu(je->nr_pending);
1149         je->nr_setting_up       = le32_to_cpu(je->nr_setting_up);
1150         je->files_open          = le32_to_cpu(je->files_open);
1151
1152         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1153                 je->m_rate[i]   = le32_to_cpu(je->m_rate[i]);
1154                 je->t_rate[i]   = le32_to_cpu(je->t_rate[i]);
1155                 je->m_iops[i]   = le32_to_cpu(je->m_iops[i]);
1156                 je->t_iops[i]   = le32_to_cpu(je->t_iops[i]);
1157                 je->rate[i]     = le32_to_cpu(je->rate[i]);
1158                 je->iops[i]     = le32_to_cpu(je->iops[i]);
1159         }
1160
1161         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
1162         je->eta_sec             = le64_to_cpu(je->eta_sec);
1163         je->nr_threads          = le32_to_cpu(je->nr_threads);
1164         je->is_pow2             = le32_to_cpu(je->is_pow2);
1165         je->unit_base           = le32_to_cpu(je->unit_base);
1166 }
1167
1168 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
1169 {
1170         int i;
1171
1172         dst->nr_running         += je->nr_running;
1173         dst->nr_ramp            += je->nr_ramp;
1174         dst->nr_pending         += je->nr_pending;
1175         dst->nr_setting_up      += je->nr_setting_up;
1176         dst->files_open         += je->files_open;
1177
1178         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1179                 dst->m_rate[i]  += je->m_rate[i];
1180                 dst->t_rate[i]  += je->t_rate[i];
1181                 dst->m_iops[i]  += je->m_iops[i];
1182                 dst->t_iops[i]  += je->t_iops[i];
1183                 dst->rate[i]    += je->rate[i];
1184                 dst->iops[i]    += je->iops[i];
1185         }
1186
1187         dst->elapsed_sec        += je->elapsed_sec;
1188
1189         if (je->eta_sec > dst->eta_sec)
1190                 dst->eta_sec = je->eta_sec;
1191
1192         dst->nr_threads         += je->nr_threads;
1193
1194         /*
1195          * This wont be correct for multiple strings, but at least it
1196          * works for the basic cases.
1197          */
1198         strcpy((char *) dst->run_str, (char *) je->run_str);
1199 }
1200
1201 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
1202 {
1203         struct fio_net_cmd_reply *reply = NULL;
1204         struct flist_head *entry;
1205
1206         flist_for_each(entry, &client->cmd_list) {
1207                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1208
1209                 if (cmd->tag == (uintptr_t) reply)
1210                         break;
1211
1212                 reply = NULL;
1213         }
1214
1215         if (!reply) {
1216                 log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag);
1217                 return;
1218         }
1219
1220         flist_del(&reply->list);
1221         cmd->tag = reply->saved_tag;
1222         free(reply);
1223 }
1224
1225 int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
1226 {
1227         do {
1228                 struct fio_net_cmd_reply *reply = NULL;
1229                 struct flist_head *entry;
1230
1231                 flist_for_each(entry, &client->cmd_list) {
1232                         reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1233
1234                         if (tag == (uintptr_t) reply)
1235                                 break;
1236
1237                         reply = NULL;
1238                 }
1239
1240                 if (!reply)
1241                         break;
1242
1243                 usleep(1000);
1244         } while (1);
1245
1246         return 0;
1247 }
1248
1249 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
1250 {
1251         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1252         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
1253
1254         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
1255
1256         assert(client->eta_in_flight == eta);
1257
1258         client->eta_in_flight = NULL;
1259         flist_del_init(&client->eta_list);
1260         client->eta_timeouts = 0;
1261
1262         if (client->ops->jobs_eta)
1263                 client->ops->jobs_eta(client, je);
1264
1265         fio_client_sum_jobs_eta(&eta->eta, je);
1266         fio_client_dec_jobs_eta(eta, client->ops->eta);
1267 }
1268
1269 static int fio_client_handle_iolog(struct fio_client *client,
1270                                    struct fio_net_cmd *cmd)
1271 {
1272         struct cmd_iolog_pdu *pdu;
1273         bool store_direct;
1274
1275         pdu = convert_iolog(cmd, &store_direct);
1276         if (!pdu) {
1277                 log_err("fio: failed converting IO log\n");
1278                 return 1;
1279         }
1280
1281         if (store_direct) {
1282                 ssize_t ret;
1283                 size_t sz;
1284                 int fd;
1285
1286                 fd = open((const char *) pdu->name,
1287                                 O_WRONLY | O_CREAT | O_TRUNC, 0644);
1288                 if (fd < 0) {
1289                         log_err("fio: open log: %s\n", strerror(errno));
1290                         return 1;
1291                 }
1292
1293                 sz = cmd->pdu_len - sizeof(*pdu);
1294                 ret = write(fd, pdu->samples, sz);
1295                 close(fd);
1296
1297                 if (ret != sz) {
1298                         log_err("fio: short write on compressed log\n");
1299                         return 1;
1300                 }
1301
1302                 return 0;
1303         } else {
1304                 FILE *f;
1305
1306                 f = fopen((const char *) pdu->name, "w");
1307                 if (!f) {
1308                         log_err("fio: fopen log: %s\n", strerror(errno));
1309                         return 1;
1310                 }
1311
1312                 flush_samples(f, pdu->samples,
1313                                 pdu->nr_samples * sizeof(struct io_sample));
1314                 fclose(f);
1315                 return 0;
1316         }
1317 }
1318
1319 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
1320 {
1321         struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
1322         const char *os, *arch;
1323         char bit[16];
1324
1325         os = fio_get_os_string(probe->os);
1326         if (!os)
1327                 os = "unknown";
1328
1329         arch = fio_get_arch_string(probe->arch);
1330         if (!arch)
1331                 os = "unknown";
1332
1333         sprintf(bit, "%d-bit", probe->bpp * 8);
1334         probe->flags = le64_to_cpu(probe->flags);
1335
1336         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n",
1337                 probe->hostname, probe->bigendian, bit, os, arch,
1338                 probe->fio_version, (unsigned long) probe->flags);
1339
1340         if (!client->name)
1341                 client->name = strdup((char *) probe->hostname);
1342 }
1343
1344 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
1345 {
1346         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1347
1348         client->state = Client_started;
1349         client->jobs = le32_to_cpu(pdu->jobs);
1350         client->nr_stat = le32_to_cpu(pdu->stat_outputs);
1351
1352         if (client->jobs) {
1353                 int i;
1354
1355                 if (client->opt_lists)
1356                         free(client->opt_lists);
1357
1358                 client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
1359                 for (i = 0; i < client->jobs; i++)
1360                         INIT_FLIST_HEAD(&client->opt_lists[i]);
1361         }
1362
1363         sum_stat_clients += client->nr_stat;
1364 }
1365
1366 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
1367 {
1368         if (client->error)
1369                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
1370 }
1371
1372 static void convert_stop(struct fio_net_cmd *cmd)
1373 {
1374         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1375
1376         pdu->error = le32_to_cpu(pdu->error);
1377 }
1378
1379 static void convert_text(struct fio_net_cmd *cmd)
1380 {
1381         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1382
1383         pdu->level      = le32_to_cpu(pdu->level);
1384         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
1385         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
1386         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
1387 }
1388
1389 static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
1390                                               struct cmd_iolog_pdu *pdu)
1391 {
1392 #ifdef CONFIG_ZLIB
1393         struct cmd_iolog_pdu *ret;
1394         z_stream stream;
1395         uint32_t nr_samples;
1396         size_t total;
1397         void *p;
1398
1399         stream.zalloc = Z_NULL;
1400         stream.zfree = Z_NULL;
1401         stream.opaque = Z_NULL;
1402         stream.avail_in = 0;
1403         stream.next_in = Z_NULL;
1404
1405         if (inflateInit(&stream) != Z_OK)
1406                 return NULL;
1407
1408         /*
1409          * Get header first, it's not compressed
1410          */
1411         nr_samples = le64_to_cpu(pdu->nr_samples);
1412
1413         total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
1414         ret = malloc(total + sizeof(*pdu));
1415         ret->nr_samples = nr_samples;
1416
1417         memcpy(ret, pdu, sizeof(*pdu));
1418
1419         p = (void *) ret + sizeof(*pdu);
1420
1421         stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1422         stream.next_in = (void *) pdu + sizeof(*pdu);
1423         while (stream.avail_in) {
1424                 unsigned int this_chunk = 65536;
1425                 unsigned int this_len;
1426                 int err;
1427
1428                 if (this_chunk > total)
1429                         this_chunk = total;
1430
1431                 stream.avail_out = this_chunk;
1432                 stream.next_out = p;
1433                 err = inflate(&stream, Z_NO_FLUSH);
1434                 /* may be Z_OK, or Z_STREAM_END */
1435                 if (err < 0) {
1436                         log_err("fio: inflate error %d\n", err);
1437                         free(ret);
1438                         ret = NULL;
1439                         goto err;
1440                 }
1441
1442                 this_len = this_chunk - stream.avail_out;
1443                 p += this_len;
1444                 total -= this_len;
1445         }
1446
1447 err:
1448         inflateEnd(&stream);
1449         return ret;
1450 #else
1451         return NULL;
1452 #endif
1453 }
1454
1455 /*
1456  * This has been compressed on the server side, since it can be big.
1457  * Uncompress here.
1458  */
1459 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd,
1460                                            bool *store_direct)
1461 {
1462         struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1463         struct cmd_iolog_pdu *ret;
1464         uint64_t i;
1465         int compressed;
1466         void *samples;
1467
1468         *store_direct = false;
1469
1470         /*
1471          * Convert if compressed and we support it. If it's not
1472          * compressed, we need not do anything.
1473          */
1474         compressed = le32_to_cpu(pdu->compressed);
1475         if (compressed == XMIT_COMPRESSED) {
1476 #ifndef CONFIG_ZLIB
1477                 log_err("fio: server sent compressed data by mistake\n");
1478                 return NULL;
1479 #endif
1480                 ret = convert_iolog_gz(cmd, pdu);
1481                 if (!ret) {
1482                         log_err("fio: failed decompressing log\n");
1483                         return NULL;
1484                 }
1485         } else if (compressed == STORE_COMPRESSED) {
1486                 *store_direct = true;
1487                 ret = pdu;
1488         } else
1489                 ret = pdu;
1490
1491         ret->nr_samples         = le64_to_cpu(ret->nr_samples);
1492         ret->thread_number      = le32_to_cpu(ret->thread_number);
1493         ret->log_type           = le32_to_cpu(ret->log_type);
1494         ret->compressed         = le32_to_cpu(ret->compressed);
1495         ret->log_offset         = le32_to_cpu(ret->log_offset);
1496
1497         if (*store_direct)
1498                 return ret;
1499
1500         samples = &ret->samples[0];
1501         for (i = 0; i < ret->nr_samples; i++) {
1502                 struct io_sample *s;
1503
1504                 s = __get_sample(samples, ret->log_offset, i);
1505                 s->time         = le64_to_cpu(s->time);
1506                 s->val          = le64_to_cpu(s->val);
1507                 s->__ddir       = le32_to_cpu(s->__ddir);
1508                 s->bs           = le32_to_cpu(s->bs);
1509
1510                 if (ret->log_offset) {
1511                         struct io_sample_offset *so = (void *) s;
1512
1513                         so->offset = le64_to_cpu(so->offset);
1514                 }
1515         }
1516
1517         return ret;
1518 }
1519
1520 static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep,
1521                            size_t size, uint64_t tag)
1522 {
1523         rep->error = cpu_to_le32(rep->error);
1524         fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL);
1525 }
1526
1527 static int fio_send_file(struct fio_client *client, struct cmd_sendfile *pdu,
1528                          uint64_t tag)
1529 {
1530         struct cmd_sendfile_reply *rep;
1531         struct stat sb;
1532         size_t size;
1533         int fd;
1534
1535         size = sizeof(*rep);
1536         rep = malloc(size);
1537
1538         if (stat((char *)pdu->path, &sb) < 0) {
1539 fail:
1540                 rep->error = errno;
1541                 sendfile_reply(client->fd, rep, size, tag);
1542                 free(rep);
1543                 return 1;
1544         }
1545
1546         size += sb.st_size;
1547         rep = realloc(rep, size);
1548         rep->size = cpu_to_le32((uint32_t) sb.st_size);
1549
1550         fd = open((char *)pdu->path, O_RDONLY);
1551         if (fd == -1 )
1552                 goto fail;
1553
1554         rep->error = read_data(fd, &rep->data, sb.st_size);
1555         sendfile_reply(client->fd, rep, size, tag);
1556         free(rep);
1557         close(fd);
1558         return 0;
1559 }
1560
1561 int fio_handle_client(struct fio_client *client)
1562 {
1563         struct client_ops *ops = client->ops;
1564         struct fio_net_cmd *cmd;
1565         int size;
1566
1567         dprint(FD_NET, "client: handle %s\n", client->hostname);
1568
1569         cmd = fio_net_recv_cmd(client->fd, true);
1570         if (!cmd)
1571                 return 0;
1572
1573         dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1574                 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1575
1576         switch (cmd->opcode) {
1577         case FIO_NET_CMD_QUIT:
1578                 if (ops->quit)
1579                         ops->quit(client, cmd);
1580                 remove_client(client);
1581                 break;
1582         case FIO_NET_CMD_TEXT:
1583                 convert_text(cmd);
1584                 ops->text(client, cmd);
1585                 break;
1586         case FIO_NET_CMD_DU: {
1587                 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1588
1589                 convert_dus(&du->dus);
1590                 convert_agg(&du->agg);
1591
1592                 ops->disk_util(client, cmd);
1593                 break;
1594                 }
1595         case FIO_NET_CMD_TS: {
1596                 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1597
1598                 dprint(FD_NET, "client: ts->ss_state = %u\n", (unsigned int) le32_to_cpu(p->ts.ss_state));
1599                 if (le32_to_cpu(p->ts.ss_state) & __FIO_SS_DATA) {
1600                         dprint(FD_NET, "client: received steadystate ring buffers\n");
1601
1602                         size = le64_to_cpu(p->ts.ss_dur);
1603                         p->ts.ss_iops_data = (uint64_t *) ((struct cmd_ts_pdu *)cmd->payload + 1);
1604                         p->ts.ss_bw_data = p->ts.ss_iops_data + size;
1605                 }
1606
1607                 convert_ts(&p->ts, &p->ts);
1608                 convert_gs(&p->rs, &p->rs);
1609
1610                 ops->thread_status(client, cmd);
1611                 break;
1612                 }
1613         case FIO_NET_CMD_GS: {
1614                 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1615
1616                 convert_gs(gs, gs);
1617
1618                 ops->group_stats(client, cmd);
1619                 break;
1620                 }
1621         case FIO_NET_CMD_ETA: {
1622                 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1623
1624                 remove_reply_cmd(client, cmd);
1625                 convert_jobs_eta(je);
1626                 handle_eta(client, cmd);
1627                 break;
1628                 }
1629         case FIO_NET_CMD_PROBE:
1630                 remove_reply_cmd(client, cmd);
1631                 ops->probe(client, cmd);
1632                 break;
1633         case FIO_NET_CMD_SERVER_START:
1634                 client->state = Client_running;
1635                 if (ops->job_start)
1636                         ops->job_start(client, cmd);
1637                 break;
1638         case FIO_NET_CMD_START: {
1639                 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1640
1641                 pdu->jobs = le32_to_cpu(pdu->jobs);
1642                 ops->start(client, cmd);
1643                 break;
1644                 }
1645         case FIO_NET_CMD_STOP: {
1646                 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1647
1648                 convert_stop(cmd);
1649                 client->state = Client_stopped;
1650                 client->error = le32_to_cpu(pdu->error);
1651                 client->signal = le32_to_cpu(pdu->signal);
1652                 ops->stop(client, cmd);
1653                 break;
1654                 }
1655         case FIO_NET_CMD_ADD_JOB: {
1656                 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1657
1658                 client->thread_number = le32_to_cpu(pdu->thread_number);
1659                 client->groupid = le32_to_cpu(pdu->groupid);
1660
1661                 if (ops->add_job)
1662                         ops->add_job(client, cmd);
1663                 break;
1664                 }
1665         case FIO_NET_CMD_IOLOG:
1666                 fio_client_handle_iolog(client, cmd);
1667                 break;
1668         case FIO_NET_CMD_UPDATE_JOB:
1669                 ops->update_job(client, cmd);
1670                 remove_reply_cmd(client, cmd);
1671                 break;
1672         case FIO_NET_CMD_VTRIGGER: {
1673                 struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
1674                 char buf[128];
1675                 int off = 0;
1676
1677                 if (aux_path) {
1678                         strcpy(buf, aux_path);
1679                         off = strlen(buf);
1680                 }
1681
1682                 __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
1683                 exec_trigger(trigger_cmd);
1684                 break;
1685                 }
1686         case FIO_NET_CMD_SENDFILE: {
1687                 struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload;
1688                 fio_send_file(client, pdu, cmd->tag);
1689                 break;
1690                 }
1691         case FIO_NET_CMD_JOB_OPT: {
1692                 handle_job_opt(client, cmd);
1693                 break;
1694         }
1695         default:
1696                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1697                 break;
1698         }
1699
1700         free(cmd);
1701         return 1;
1702 }
1703
1704 int fio_clients_send_trigger(const char *cmd)
1705 {
1706         struct flist_head *entry;
1707         struct fio_client *client;
1708         size_t slen;
1709
1710         dprint(FD_NET, "client: send vtrigger: %s\n", cmd);
1711
1712         if (!cmd)
1713                 slen = 0;
1714         else
1715                 slen = strlen(cmd);
1716
1717         flist_for_each(entry, &client_list) {
1718                 struct cmd_vtrigger_pdu *pdu;
1719
1720                 client = flist_entry(entry, struct fio_client, list);
1721
1722                 pdu = malloc(sizeof(*pdu) + slen);
1723                 pdu->len = cpu_to_le16((uint16_t) slen);
1724                 if (slen)
1725                         memcpy(pdu->cmd, cmd, slen);
1726                 fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu,
1727                                         sizeof(*pdu) + slen, NULL, NULL);
1728                 free(pdu);
1729         }
1730
1731         return 0;
1732 }
1733
1734 static void request_client_etas(struct client_ops *ops)
1735 {
1736         struct fio_client *client;
1737         struct flist_head *entry;
1738         struct client_eta *eta;
1739         int skipped = 0;
1740
1741         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1742
1743         eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
1744         eta->pending = nr_clients;
1745
1746         flist_for_each(entry, &client_list) {
1747                 client = flist_entry(entry, struct fio_client, list);
1748
1749                 if (!flist_empty(&client->eta_list)) {
1750                         skipped++;
1751                         continue;
1752                 }
1753                 if (client->state != Client_running)
1754                         continue;
1755
1756                 assert(!client->eta_in_flight);
1757                 flist_add_tail(&client->eta_list, &eta_list);
1758                 client->eta_in_flight = eta;
1759                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1760                                         (uintptr_t) eta, &client->cmd_list);
1761         }
1762
1763         while (skipped--) {
1764                 if (!fio_client_dec_jobs_eta(eta, ops->eta))
1765                         break;
1766         }
1767
1768         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1769 }
1770
1771 /*
1772  * A single SEND_ETA timeout isn't fatal. Attempt to recover.
1773  */
1774 static int handle_cmd_timeout(struct fio_client *client,
1775                               struct fio_net_cmd_reply *reply)
1776 {
1777         flist_del(&reply->list);
1778         free(reply);
1779
1780         if (reply->opcode != FIO_NET_CMD_SEND_ETA)
1781                 return 1;
1782
1783         log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
1784
1785         flist_del_init(&client->eta_list);
1786         if (client->eta_in_flight) {
1787                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
1788                 client->eta_in_flight = NULL;
1789         }
1790
1791         /*
1792          * If we fail 5 in a row, give up...
1793          */
1794         if (client->eta_timeouts++ > 5)
1795                 return 1;
1796
1797         return 0;
1798 }
1799
1800 static int client_check_cmd_timeout(struct fio_client *client,
1801                                     struct timeval *now)
1802 {
1803         struct fio_net_cmd_reply *reply;
1804         struct flist_head *entry, *tmp;
1805         int ret = 0;
1806
1807         flist_for_each_safe(entry, tmp, &client->cmd_list) {
1808                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1809
1810                 if (mtime_since(&reply->tv, now) < FIO_NET_CLIENT_TIMEOUT)
1811                         continue;
1812
1813                 if (!handle_cmd_timeout(client, reply))
1814                         continue;
1815
1816                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1817                                                 fio_server_op(reply->opcode));
1818                 ret = 1;
1819         }
1820
1821         return flist_empty(&client->cmd_list) && ret;
1822 }
1823
1824 static int fio_check_clients_timed_out(void)
1825 {
1826         struct fio_client *client;
1827         struct flist_head *entry, *tmp;
1828         struct timeval tv;
1829         int ret = 0;
1830
1831         fio_gettime(&tv, NULL);
1832
1833         flist_for_each_safe(entry, tmp, &client_list) {
1834                 client = flist_entry(entry, struct fio_client, list);
1835
1836                 if (flist_empty(&client->cmd_list))
1837                         continue;
1838
1839                 if (!client_check_cmd_timeout(client, &tv))
1840                         continue;
1841
1842                 if (client->ops->timed_out)
1843                         client->ops->timed_out(client);
1844                 else
1845                         log_err("fio: client %s timed out\n", client->hostname);
1846
1847                 client->error = ETIMEDOUT;
1848                 remove_client(client);
1849                 ret = 1;
1850         }
1851
1852         return ret;
1853 }
1854
1855 int fio_handle_clients(struct client_ops *ops)
1856 {
1857         struct pollfd *pfds;
1858         int i, ret = 0, retval = 0;
1859
1860         fio_gettime(&eta_tv, NULL);
1861
1862         pfds = malloc(nr_clients * sizeof(struct pollfd));
1863
1864         init_thread_stat(&client_ts);
1865         init_group_run_stat(&client_gs);
1866
1867         while (!exit_backend && nr_clients) {
1868                 struct flist_head *entry, *tmp;
1869                 struct fio_client *client;
1870
1871                 i = 0;
1872                 flist_for_each_safe(entry, tmp, &client_list) {
1873                         client = flist_entry(entry, struct fio_client, list);
1874
1875                         if (!client->sent_job && !client->ops->stay_connected &&
1876                             flist_empty(&client->cmd_list)) {
1877                                 remove_client(client);
1878                                 continue;
1879                         }
1880
1881                         pfds[i].fd = client->fd;
1882                         pfds[i].events = POLLIN;
1883                         i++;
1884                 }
1885
1886                 if (!nr_clients)
1887                         break;
1888
1889                 assert(i == nr_clients);
1890
1891                 do {
1892                         struct timeval tv;
1893                         int timeout;
1894
1895                         fio_gettime(&tv, NULL);
1896                         if (mtime_since(&eta_tv, &tv) >= 900) {
1897                                 request_client_etas(ops);
1898                                 memcpy(&eta_tv, &tv, sizeof(tv));
1899
1900                                 if (fio_check_clients_timed_out())
1901                                         break;
1902                         }
1903
1904                         check_trigger_file();
1905
1906                         timeout = min(100u, ops->eta_msec);
1907
1908                         ret = poll(pfds, nr_clients, timeout);
1909                         if (ret < 0) {
1910                                 if (errno == EINTR)
1911                                         continue;
1912                                 log_err("fio: poll clients: %s\n", strerror(errno));
1913                                 break;
1914                         } else if (!ret)
1915                                 continue;
1916                 } while (ret <= 0);
1917
1918                 for (i = 0; i < nr_clients; i++) {
1919                         if (!(pfds[i].revents & POLLIN))
1920                                 continue;
1921
1922                         client = find_client_by_fd(pfds[i].fd);
1923                         if (!client) {
1924                                 log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd);
1925                                 continue;
1926                         }
1927                         if (!fio_handle_client(client)) {
1928                                 log_info("client: host=%s disconnected\n",
1929                                                 client->hostname);
1930                                 remove_client(client);
1931                                 retval = 1;
1932                         } else if (client->error)
1933                                 retval = 1;
1934                         fio_put_client(client);
1935                 }
1936         }
1937
1938         fio_client_json_fini();
1939
1940         free(pfds);
1941         return retval || error_clients;
1942 }