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_object *job_opt_object = NULL;
  63 static struct json_array *clients_array = NULL;
  64 static struct json_array *du_array = NULL;
  65
  66 static int error_clients;
  67
  68 #define FIO_CLIENT_HASH_BITS    7
  69 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  70 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  71 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  72
  73 static void fio_client_add_hash(struct fio_client *client)
  74 {
  75         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  76
  77         bucket &= FIO_CLIENT_HASH_MASK;
  78         flist_add(&client->hash_list, &client_hash[bucket]);
  79 }
  80
  81 static void fio_client_remove_hash(struct fio_client *client)
  82 {
  83         if (!flist_empty(&client->hash_list))
  84                 flist_del_init(&client->hash_list);
  85 }
  86
  87 static void fio_init fio_client_hash_init(void)
  88 {
  89         int i;
  90
  91         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
  92                 INIT_FLIST_HEAD(&client_hash[i]);
  93 }
  94
  95 static int read_data(int fd, void *data, size_t size)
  96 {
  97         ssize_t ret;
  98
  99         while (size) {
 100                 ret = read(fd, data, size);
 101                 if (ret < 0) {
 102                         if (errno == EAGAIN || errno == EINTR)
 103                                 continue;
 104                         break;
 105                 } else if (!ret)
 106                         break;
 107                 else {
 108                         data += ret;
 109                         size -= ret;
 110                 }
 111         }
 112
 113         if (size)
 114                 return EAGAIN;
 115
 116         return 0;
 117 }
 118
 119 static void fio_client_json_init(void)
 120 {
 121         char time_buf[32];
 122         time_t time_p;
 123
 124         if (!(output_format & FIO_OUTPUT_JSON))
 125                 return;
 126
 127         time(&time_p);
 128         os_ctime_r((const time_t *) &time_p, time_buf, sizeof(time_buf));
 129         time_buf[strlen(time_buf) - 1] = '\0';
 130
 131         root = json_create_object();
 132         json_object_add_value_string(root, "fio version", fio_version_string);
 133         json_object_add_value_int(root, "timestamp", time_p);
 134         json_object_add_value_string(root, "time", time_buf);
 135
 136         job_opt_object = json_create_object();
 137         json_object_add_value_object(root, "global options", job_opt_object);
 138         clients_array = json_create_array();
 139         json_object_add_value_array(root, "client_stats", clients_array);
 140         du_array = json_create_array();
 141         json_object_add_value_array(root, "disk_util", du_array);
 142 }
 143
 144 static void fio_client_json_fini(void)
 145 {
 146         if (!(output_format & FIO_OUTPUT_JSON))
 147                 return;
 148
 149         log_info("\n");
 150         json_print_object(root, NULL);
 151         log_info("\n");
 152         json_free_object(root);
 153         root = NULL;
 154         clients_array = NULL;
 155         du_array = NULL;
 156 }
 157
 158 static struct fio_client *find_client_by_fd(int fd)
 159 {
 160         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
 161         struct fio_client *client;
 162         struct flist_head *entry;
 163
 164         flist_for_each(entry, &client_hash[bucket]) {
 165                 client = flist_entry(entry, struct fio_client, hash_list);
 166
 167                 if (client->fd == fd) {
 168                         client->refs++;
 169                         return client;
 170                 }
 171         }
 172
 173         return NULL;
 174 }
 175
 176 void fio_put_client(struct fio_client *client)
 177 {
 178         if (--client->refs)
 179                 return;
 180
 181         free(client->hostname);
 182         if (client->argv)
 183                 free(client->argv);
 184         if (client->name)
 185                 free(client->name);
 186         while (client->nr_files) {
 187                 struct client_file *cf = &client->files[--client->nr_files];
 188
 189                 free(cf->file);
 190         }
 191         if (client->files)
 192                 free(client->files);
 193         if (client->opt_lists)
 194                 free(client->opt_lists);
 195
 196         if (!client->did_stat)
 197                 sum_stat_clients--;
 198
 199         if (client->error)
 200                 error_clients++;
 201
 202         free(client);
 203 }
 204
 205 static int fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
 206 {
 207         if (!--eta->pending) {
 208                 eta_fn(&eta->eta);
 209                 free(eta);
 210                 return 0;
 211         }
 212
 213         return 1;
 214 }
 215
 216 static void remove_client(struct fio_client *client)
 217 {
 218         assert(client->refs);
 219
 220         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 221
 222         if (!flist_empty(&client->list))
 223                 flist_del_init(&client->list);
 224
 225         fio_client_remove_hash(client);
 226
 227         if (!flist_empty(&client->eta_list)) {
 228                 flist_del_init(&client->eta_list);
 229                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
 230         }
 231
 232         close(client->fd);
 233         client->fd = -1;
 234
 235         if (client->ops->removed)
 236                 client->ops->removed(client);
 237
 238         nr_clients--;
 239         fio_put_client(client);
 240 }
 241
 242 struct fio_client *fio_get_client(struct fio_client *client)
 243 {
 244         client->refs++;
 245         return client;
 246 }
 247
 248 static void __fio_client_add_cmd_option(struct fio_client *client,
 249                                         const char *opt)
 250 {
 251         int index;
 252
 253         index = client->argc++;
 254         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 255         client->argv[index] = strdup(opt);
 256         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 257 }
 258
 259 void fio_client_add_cmd_option(void *cookie, const char *opt)
 260 {
 261         struct fio_client *client = cookie;
 262         struct flist_head *entry;
 263
 264         if (!client || !opt)
 265                 return;
 266
 267         __fio_client_add_cmd_option(client, opt);
 268
 269         /*
 270          * Duplicate arguments to shared client group
 271          */
 272         flist_for_each(entry, &arg_list) {
 273                 client = flist_entry(entry, struct fio_client, arg_list);
 274
 275                 __fio_client_add_cmd_option(client, opt);
 276         }
 277 }
 278
 279 struct fio_client *fio_client_add_explicit(struct client_ops *ops,
 280                                            const char *hostname, int type,
 281                                            int port)
 282 {
 283         struct fio_client *client;
 284
 285         client = malloc(sizeof(*client));
 286         memset(client, 0, sizeof(*client));
 287
 288         INIT_FLIST_HEAD(&client->list);
 289         INIT_FLIST_HEAD(&client->hash_list);
 290         INIT_FLIST_HEAD(&client->arg_list);
 291         INIT_FLIST_HEAD(&client->eta_list);
 292         INIT_FLIST_HEAD(&client->cmd_list);
 293
 294         client->hostname = strdup(hostname);
 295
 296         if (type == Fio_client_socket)
 297                 client->is_sock = 1;
 298         else {
 299                 int ipv6;
 300
 301                 ipv6 = type == Fio_client_ipv6;
 302                 if (fio_server_parse_host(hostname, ipv6,
 303                                                 &client->addr.sin_addr,
 304                                                 &client->addr6.sin6_addr))
 305                         goto err;
 306
 307                 client->port = port;
 308         }
 309
 310         client->fd = -1;
 311         client->ops = ops;
 312         client->refs = 1;
 313         client->type = ops->client_type;
 314
 315         __fio_client_add_cmd_option(client, "fio");
 316
 317         flist_add(&client->list, &client_list);
 318         nr_clients++;
 319         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 320         return client;
 321 err:
 322         free(client);
 323         return NULL;
 324 }
 325
 326 int fio_client_add_ini_file(void *cookie, const char *ini_file, int remote)
 327 {
 328         struct fio_client *client = cookie;
 329         struct client_file *cf;
 330         size_t new_size;
 331         void *new_files;
 332
 333         if (!client)
 334                 return 1;
 335
 336         dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file);
 337
 338         new_size = (client->nr_files + 1) * sizeof(struct client_file);
 339         new_files = realloc(client->files, new_size);
 340         if (!new_files)
 341                 return 1;
 342
 343         client->files = new_files;
 344         cf = &client->files[client->nr_files];
 345         cf->file = strdup(ini_file);
 346         cf->remote = remote;
 347         client->nr_files++;
 348         return 0;
 349 }
 350
 351 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
 352 {
 353         struct fio_client *existing = *cookie;
 354         struct fio_client *client;
 355
 356         if (existing) {
 357                 /*
 358                  * We always add our "exec" name as the option, hence 1
 359                  * means empty.
 360                  */
 361                 if (existing->argc == 1)
 362                         flist_add_tail(&existing->arg_list, &arg_list);
 363                 else {
 364                         while (!flist_empty(&arg_list))
 365                                 flist_del_init(arg_list.next);
 366                 }
 367         }
 368
 369         client = malloc(sizeof(*client));
 370         memset(client, 0, sizeof(*client));
 371
 372         INIT_FLIST_HEAD(&client->list);
 373         INIT_FLIST_HEAD(&client->hash_list);
 374         INIT_FLIST_HEAD(&client->arg_list);
 375         INIT_FLIST_HEAD(&client->eta_list);
 376         INIT_FLIST_HEAD(&client->cmd_list);
 377
 378         if (fio_server_parse_string(hostname, &client->hostname,
 379                                         &client->is_sock, &client->port,
 380                                         &client->addr.sin_addr,
 381                                         &client->addr6.sin6_addr,
 382                                         &client->ipv6))
 383                 return -1;
 384
 385         client->fd = -1;
 386         client->ops = ops;
 387         client->refs = 1;
 388         client->type = ops->client_type;
 389
 390         __fio_client_add_cmd_option(client, "fio");
 391
 392         flist_add(&client->list, &client_list);
 393         nr_clients++;
 394         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 395         *cookie = client;
 396         return 0;
 397 }
 398
 399 static const char *server_name(struct fio_client *client, char *buf,
 400                                size_t bufsize)
 401 {
 402         const char *from;
 403
 404         if (client->ipv6)
 405                 from = inet_ntop(AF_INET6, (struct sockaddr *) &client->addr6.sin6_addr, buf, bufsize);
 406         else if (client->is_sock)
 407                 from = "sock";
 408         else
 409                 from = inet_ntop(AF_INET, (struct sockaddr *) &client->addr.sin_addr, buf, bufsize);
 410
 411         return from;
 412 }
 413
 414 static void probe_client(struct fio_client *client)
 415 {
 416         struct cmd_client_probe_pdu pdu;
 417         const char *sname;
 418         uint64_t tag;
 419         char buf[64];
 420
 421         dprint(FD_NET, "client: send probe\n");
 422
 423 #ifdef CONFIG_ZLIB
 424         pdu.flags = __le64_to_cpu(FIO_PROBE_FLAG_ZLIB);
 425 #else
 426         pdu.flags = 0;
 427 #endif
 428
 429         sname = server_name(client, buf, sizeof(buf));
 430         memset(pdu.server, 0, sizeof(pdu.server));
 431         strncpy((char *) pdu.server, sname, sizeof(pdu.server) - 1);
 432
 433         fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list);
 434 }
 435
 436 static int fio_client_connect_ip(struct fio_client *client)
 437 {
 438         struct sockaddr *addr;
 439         socklen_t socklen;
 440         int fd, domain;
 441
 442         if (client->ipv6) {
 443                 client->addr6.sin6_family = AF_INET6;
 444                 client->addr6.sin6_port = htons(client->port);
 445                 domain = AF_INET6;
 446                 addr = (struct sockaddr *) &client->addr6;
 447                 socklen = sizeof(client->addr6);
 448         } else {
 449                 client->addr.sin_family = AF_INET;
 450                 client->addr.sin_port = htons(client->port);
 451                 domain = AF_INET;
 452                 addr = (struct sockaddr *) &client->addr;
 453                 socklen = sizeof(client->addr);
 454         }
 455
 456         fd = socket(domain, SOCK_STREAM, 0);
 457         if (fd < 0) {
 458                 int ret = -errno;
 459
 460                 log_err("fio: socket: %s\n", strerror(errno));
 461                 return ret;
 462         }
 463
 464         if (connect(fd, addr, socklen) < 0) {
 465                 int ret = -errno;
 466
 467                 log_err("fio: connect: %s\n", strerror(errno));
 468                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 469                                                                 client->port);
 470                 close(fd);
 471                 return ret;
 472         }
 473
 474         return fd;
 475 }
 476
 477 static int fio_client_connect_sock(struct fio_client *client)
 478 {
 479         struct sockaddr_un *addr = &client->addr_un;
 480         socklen_t len;
 481         int fd;
 482
 483         memset(addr, 0, sizeof(*addr));
 484         addr->sun_family = AF_UNIX;
 485         strncpy(addr->sun_path, client->hostname, sizeof(addr->sun_path) - 1);
 486
 487         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 488         if (fd < 0) {
 489                 int ret = -errno;
 490
 491                 log_err("fio: socket: %s\n", strerror(errno));
 492                 return ret;
 493         }
 494
 495         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 496         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 497                 int ret = -errno;
 498
 499                 log_err("fio: connect; %s\n", strerror(errno));
 500                 close(fd);
 501                 return ret;
 502         }
 503
 504         return fd;
 505 }
 506
 507 int fio_client_connect(struct fio_client *client)
 508 {
 509         int fd;
 510
 511         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 512
 513         if (client->is_sock)
 514                 fd = fio_client_connect_sock(client);
 515         else
 516                 fd = fio_client_connect_ip(client);
 517
 518         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 519
 520         if (fd < 0)
 521                 return fd;
 522
 523         client->fd = fd;
 524         fio_client_add_hash(client);
 525         client->state = Client_connected;
 526
 527         probe_client(client);
 528         return 0;
 529 }
 530
 531 int fio_client_terminate(struct fio_client *client)
 532 {
 533         return fio_net_send_quit(client->fd);
 534 }
 535
 536 void fio_clients_terminate(void)
 537 {
 538         struct flist_head *entry;
 539         struct fio_client *client;
 540
 541         dprint(FD_NET, "client: terminate clients\n");
 542
 543         flist_for_each(entry, &client_list) {
 544                 client = flist_entry(entry, struct fio_client, list);
 545                 fio_client_terminate(client);
 546         }
 547 }
 548
 549 static void sig_int(int sig)
 550 {
 551         dprint(FD_NET, "client: got signal %d\n", sig);
 552         fio_clients_terminate();
 553 }
 554
 555 static void client_signal_handler(void)
 556 {
 557         struct sigaction act;
 558
 559         memset(&act, 0, sizeof(act));
 560         act.sa_handler = sig_int;
 561         act.sa_flags = SA_RESTART;
 562         sigaction(SIGINT, &act, NULL);
 563
 564         memset(&act, 0, sizeof(act));
 565         act.sa_handler = sig_int;
 566         act.sa_flags = SA_RESTART;
 567         sigaction(SIGTERM, &act, NULL);
 568
 569 /* Windows uses SIGBREAK as a quit signal from other applications */
 570 #ifdef WIN32
 571         memset(&act, 0, sizeof(act));
 572         act.sa_handler = sig_int;
 573         act.sa_flags = SA_RESTART;
 574         sigaction(SIGBREAK, &act, NULL);
 575 #endif
 576
 577         memset(&act, 0, sizeof(act));
 578         act.sa_handler = sig_show_status;
 579         act.sa_flags = SA_RESTART;
 580         sigaction(SIGUSR1, &act, NULL);
 581 }
 582
 583 static int send_client_cmd_line(struct fio_client *client)
 584 {
 585         struct cmd_single_line_pdu *cslp;
 586         struct cmd_line_pdu *clp;
 587         unsigned long offset;
 588         unsigned int *lens;
 589         void *pdu;
 590         size_t mem;
 591         int i, ret;
 592
 593         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 594
 595         lens = malloc(client->argc * sizeof(unsigned int));
 596
 597         /*
 598          * Find out how much mem we need
 599          */
 600         for (i = 0, mem = 0; i < client->argc; i++) {
 601                 lens[i] = strlen(client->argv[i]) + 1;
 602                 mem += lens[i];
 603         }
 604
 605         /*
 606          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 607          */
 608         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 609
 610         pdu = malloc(mem);
 611         clp = pdu;
 612         offset = sizeof(*clp);
 613
 614         for (i = 0; i < client->argc; i++) {
 615                 uint16_t arg_len = lens[i];
 616
 617                 cslp = pdu + offset;
 618                 strcpy((char *) cslp->text, client->argv[i]);
 619                 cslp->len = cpu_to_le16(arg_len);
 620                 offset += sizeof(*cslp) + arg_len;
 621         }
 622
 623         free(lens);
 624         clp->lines = cpu_to_le16(client->argc);
 625         clp->client_type = __cpu_to_le16(client->type);
 626         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, NULL, NULL);
 627         free(pdu);
 628         return ret;
 629 }
 630
 631 int fio_clients_connect(void)
 632 {
 633         struct fio_client *client;
 634         struct flist_head *entry, *tmp;
 635         int ret;
 636
 637 #ifdef WIN32
 638         WSADATA wsd;
 639         WSAStartup(MAKEWORD(2, 2), &wsd);
 640 #endif
 641
 642         dprint(FD_NET, "client: connect all\n");
 643
 644         client_signal_handler();
 645
 646         flist_for_each_safe(entry, tmp, &client_list) {
 647                 client = flist_entry(entry, struct fio_client, list);
 648
 649                 ret = fio_client_connect(client);
 650                 if (ret) {
 651                         remove_client(client);
 652                         continue;
 653                 }
 654
 655                 if (client->argc > 1)
 656                         send_client_cmd_line(client);
 657         }
 658
 659         return !nr_clients;
 660 }
 661
 662 int fio_start_client(struct fio_client *client)
 663 {
 664         dprint(FD_NET, "client: start %s\n", client->hostname);
 665         return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
 666 }
 667
 668 int fio_start_all_clients(void)
 669 {
 670         struct fio_client *client;
 671         struct flist_head *entry, *tmp;
 672         int ret;
 673
 674         dprint(FD_NET, "client: start all\n");
 675
 676         fio_client_json_init();
 677
 678         flist_for_each_safe(entry, tmp, &client_list) {
 679                 client = flist_entry(entry, struct fio_client, list);
 680
 681                 ret = fio_start_client(client);
 682                 if (ret) {
 683                         remove_client(client);
 684                         continue;
 685                 }
 686         }
 687
 688         return flist_empty(&client_list);
 689 }
 690
 691 static int __fio_client_send_remote_ini(struct fio_client *client,
 692                                         const char *filename)
 693 {
 694         struct cmd_load_file_pdu *pdu;
 695         size_t p_size;
 696         int ret;
 697
 698         dprint(FD_NET, "send remote ini %s to %s\n", filename, client->hostname);
 699
 700         p_size = sizeof(*pdu) + strlen(filename) + 1;
 701         pdu = malloc(p_size);
 702         memset(pdu, 0, p_size);
 703         pdu->name_len = strlen(filename);
 704         strcpy((char *) pdu->file, filename);
 705         pdu->client_type = cpu_to_le16((uint16_t) client->type);
 706
 707         client->sent_job = 1;
 708         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_LOAD_FILE, pdu, p_size,NULL, NULL);
 709         free(pdu);
 710         return ret;
 711 }
 712
 713 /*
 714  * Send file contents to server backend. We could use sendfile(), but to remain
 715  * more portable lets just read/write the darn thing.
 716  */
 717 static int __fio_client_send_local_ini(struct fio_client *client,
 718                                        const char *filename)
 719 {
 720         struct cmd_job_pdu *pdu;
 721         size_t p_size;
 722         struct stat sb;
 723         char *p;
 724         void *buf;
 725         off_t len;
 726         int fd, ret;
 727
 728         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 729
 730         fd = open(filename, O_RDONLY);
 731         if (fd < 0) {
 732                 ret = -errno;
 733                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 734                 return ret;
 735         }
 736
 737         if (fstat(fd, &sb) < 0) {
 738                 ret = -errno;
 739                 log_err("fio: job file stat: %s\n", strerror(errno));
 740                 close(fd);
 741                 return ret;
 742         }
 743
 744         p_size = sb.st_size + sizeof(*pdu);
 745         pdu = malloc(p_size);
 746         buf = pdu->buf;
 747
 748         len = sb.st_size;
 749         p = buf;
 750         if (read_data(fd, p, len)) {
 751                 log_err("fio: failed reading job file %s\n", filename);
 752                 close(fd);
 753                 free(pdu);
 754                 return 1;
 755         }
 756
 757         pdu->buf_len = __cpu_to_le32(sb.st_size);
 758         pdu->client_type = cpu_to_le32(client->type);
 759
 760         client->sent_job = 1;
 761         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, NULL, NULL);
 762         free(pdu);
 763         close(fd);
 764         return ret;
 765 }
 766
 767 int fio_client_send_ini(struct fio_client *client, const char *filename,
 768                         int remote)
 769 {
 770         int ret;
 771
 772         if (!remote)
 773                 ret = __fio_client_send_local_ini(client, filename);
 774         else
 775                 ret = __fio_client_send_remote_ini(client, filename);
 776
 777         if (!ret)
 778                 client->sent_job = 1;
 779
 780         return ret;
 781 }
 782
 783 static int fio_client_send_cf(struct fio_client *client,
 784                               struct client_file *cf)
 785 {
 786         return fio_client_send_ini(client, cf->file, cf->remote);
 787 }
 788
 789 int fio_clients_send_ini(const char *filename)
 790 {
 791         struct fio_client *client;
 792         struct flist_head *entry, *tmp;
 793
 794         flist_for_each_safe(entry, tmp, &client_list) {
 795                 client = flist_entry(entry, struct fio_client, list);
 796
 797                 if (client->nr_files) {
 798                         int i;
 799
 800                         for (i = 0; i < client->nr_files; i++) {
 801                                 struct client_file *cf;
 802
 803                                 cf = &client->files[i];
 804
 805                                 if (fio_client_send_cf(client, cf)) {
 806                                         remove_client(client);
 807                                         break;
 808                                 }
 809                         }
 810                 }
 811                 if (client->sent_job)
 812                         continue;
 813                 if (!filename || fio_client_send_ini(client, filename, 0))
 814                         remove_client(client);
 815         }
 816
 817         return !nr_clients;
 818 }
 819
 820 int fio_client_update_options(struct fio_client *client,
 821                               struct thread_options *o, uint64_t *tag)
 822 {
 823         struct cmd_add_job_pdu pdu;
 824
 825         pdu.thread_number = cpu_to_le32(client->thread_number);
 826         pdu.groupid = cpu_to_le32(client->groupid);
 827         convert_thread_options_to_net(&pdu.top, o);
 828
 829         return fio_net_send_cmd(client->fd, FIO_NET_CMD_UPDATE_JOB, &pdu, sizeof(pdu), tag, &client->cmd_list);
 830 }
 831
 832 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 833 {
 834         dst->max_val    = le64_to_cpu(src->max_val);
 835         dst->min_val    = le64_to_cpu(src->min_val);
 836         dst->samples    = le64_to_cpu(src->samples);
 837
 838         /*
 839          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 840          */
 841         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 842         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 843 }
 844
 845 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 846 {
 847         int i, j;
 848
 849         dst->error              = le32_to_cpu(src->error);
 850         dst->thread_number      = le32_to_cpu(src->thread_number);
 851         dst->groupid            = le32_to_cpu(src->groupid);
 852         dst->pid                = le32_to_cpu(src->pid);
 853         dst->members            = le32_to_cpu(src->members);
 854         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
 855
 856         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 857                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 858                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 859                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 860                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 861         }
 862
 863         dst->usr_time           = le64_to_cpu(src->usr_time);
 864         dst->sys_time           = le64_to_cpu(src->sys_time);
 865         dst->ctx                = le64_to_cpu(src->ctx);
 866         dst->minf               = le64_to_cpu(src->minf);
 867         dst->majf               = le64_to_cpu(src->majf);
 868         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 869         dst->percentile_precision = le64_to_cpu(src->percentile_precision);
 870
 871         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 872                 fio_fp64_t *fps = &src->percentile_list[i];
 873                 fio_fp64_t *fpd = &dst->percentile_list[i];
 874
 875                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 876         }
 877
 878         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 879                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 880                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 881                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 882         }
 883
 884         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 885                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 886                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 887         }
 888
 889         for (i = 0; i < DDIR_RWDIR_CNT; i++)
 890                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 891                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 892
 893         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 894                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 895                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 896                 dst->drop_io_u[i]       = le64_to_cpu(src->drop_io_u[i]);
 897         }
 898
 899         dst->total_submit       = le64_to_cpu(src->total_submit);
 900         dst->total_complete     = le64_to_cpu(src->total_complete);
 901
 902         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 903                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 904                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 905         }
 906
 907         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 908         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 909         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 910         dst->first_error        = le32_to_cpu(src->first_error);
 911         dst->kb_base            = le32_to_cpu(src->kb_base);
 912         dst->unit_base          = le32_to_cpu(src->unit_base);
 913
 914         dst->latency_depth      = le32_to_cpu(src->latency_depth);
 915         dst->latency_target     = le64_to_cpu(src->latency_target);
 916         dst->latency_window     = le64_to_cpu(src->latency_window);
 917         dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i));
 918
 919         dst->nr_block_infos     = le64_to_cpu(src->nr_block_infos);
 920         for (i = 0; i < dst->nr_block_infos; i++)
 921                 dst->block_infos[i] = le32_to_cpu(src->block_infos[i]);
 922 }
 923
 924 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 925 {
 926         int i;
 927
 928         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 929                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 930                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 931                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 932                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 933                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 934                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 935         }
 936
 937         dst->kb_base    = le32_to_cpu(src->kb_base);
 938         dst->unit_base  = le32_to_cpu(src->unit_base);
 939         dst->groupid    = le32_to_cpu(src->groupid);
 940         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
 941 }
 942
 943 static void json_object_add_client_info(struct json_object *obj,
 944                                         struct fio_client *client)
 945 {
 946         const char *hostname = client->hostname ? client->hostname : "";
 947
 948         json_object_add_value_string(obj, "hostname", hostname);
 949         json_object_add_value_int(obj, "port", client->port);
 950 }
 951
 952 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
 953 {
 954         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 955         struct flist_head *opt_list = NULL;
 956         struct json_object *tsobj;
 957
 958         if (client->opt_lists && p->ts.thread_number <= client->jobs)
 959                 opt_list = &client->opt_lists[p->ts.thread_number - 1];
 960
 961         tsobj = show_thread_status(&p->ts, &p->rs, opt_list, NULL);
 962         client->did_stat = 1;
 963         if (tsobj) {
 964                 json_object_add_client_info(tsobj, client);
 965                 json_array_add_value_object(clients_array, tsobj);
 966         }
 967
 968         if (sum_stat_clients <= 1)
 969                 return;
 970
 971         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1);
 972         sum_group_stats(&client_gs, &p->rs);
 973
 974         client_ts.members++;
 975         client_ts.thread_number = p->ts.thread_number;
 976         client_ts.groupid = p->ts.groupid;
 977         client_ts.unified_rw_rep = p->ts.unified_rw_rep;
 978
 979         if (++sum_stat_nr == sum_stat_clients) {
 980                 strcpy(client_ts.name, "All clients");
 981                 tsobj = show_thread_status(&client_ts, &client_gs, NULL, NULL);
 982                 if (tsobj) {
 983                         json_object_add_client_info(tsobj, client);
 984                         json_array_add_value_object(clients_array, tsobj);
 985                 }
 986         }
 987 }
 988
 989 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
 990 {
 991         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 992
 993         if (output_format & FIO_OUTPUT_NORMAL)
 994                 show_group_stats(gs, NULL);
 995 }
 996
 997 static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd)
 998 {
 999         struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload;
1000         struct print_option *p;
1001
1002         pdu->global = le16_to_cpu(pdu->global);
1003         pdu->groupid = le16_to_cpu(pdu->groupid);
1004
1005         p = malloc(sizeof(*p));
1006         p->name = strdup((char *) pdu->name);
1007         if (pdu->value[0] != '\0')
1008                 p->value = strdup((char *) pdu->value);
1009         else
1010                 p->value = NULL;
1011
1012         if (pdu->global) {
1013                 const char *pos = "";
1014
1015                 if (p->value)
1016                         pos = p->value;
1017
1018                 json_object_add_value_string(job_opt_object, p->name, pos);
1019         } else if (client->opt_lists) {
1020                 struct flist_head *opt_list = &client->opt_lists[pdu->groupid];
1021
1022                 flist_add_tail(&p->list, opt_list);
1023         }
1024 }
1025
1026 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
1027 {
1028         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1029         const char *buf = (const char *) pdu->buf;
1030         const char *name;
1031         int fio_unused ret;
1032
1033         name = client->name ? client->name : client->hostname;
1034
1035         if (!client->skip_newline)
1036                 fprintf(f_out, "<%s> ", name);
1037         ret = fwrite(buf, pdu->buf_len, 1, f_out);
1038         fflush(f_out);
1039         client->skip_newline = strchr(buf, '\n') == NULL;
1040 }
1041
1042 static void convert_agg(struct disk_util_agg *agg)
1043 {
1044         int i;
1045
1046         for (i = 0; i < 2; i++) {
1047                 agg->ios[i]     = le64_to_cpu(agg->ios[i]);
1048                 agg->merges[i]  = le64_to_cpu(agg->merges[i]);
1049                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
1050                 agg->ticks[i]   = le64_to_cpu(agg->ticks[i]);
1051         }
1052
1053         agg->io_ticks           = le64_to_cpu(agg->io_ticks);
1054         agg->time_in_queue      = le64_to_cpu(agg->time_in_queue);
1055         agg->slavecount         = le32_to_cpu(agg->slavecount);
1056         agg->max_util.u.f       = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i));
1057 }
1058
1059 static void convert_dus(struct disk_util_stat *dus)
1060 {
1061         int i;
1062
1063         for (i = 0; i < 2; i++) {
1064                 dus->s.ios[i]           = le64_to_cpu(dus->s.ios[i]);
1065                 dus->s.merges[i]        = le64_to_cpu(dus->s.merges[i]);
1066                 dus->s.sectors[i]       = le64_to_cpu(dus->s.sectors[i]);
1067                 dus->s.ticks[i]         = le64_to_cpu(dus->s.ticks[i]);
1068         }
1069
1070         dus->s.io_ticks         = le64_to_cpu(dus->s.io_ticks);
1071         dus->s.time_in_queue    = le64_to_cpu(dus->s.time_in_queue);
1072         dus->s.msec             = le64_to_cpu(dus->s.msec);
1073 }
1074
1075 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
1076 {
1077         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1078
1079         if (!client->disk_stats_shown) {
1080                 client->disk_stats_shown = 1;
1081                 log_info("\nDisk stats (read/write):\n");
1082         }
1083
1084         if (output_format & FIO_OUTPUT_JSON) {
1085                 struct json_object *duobj;
1086                 json_array_add_disk_util(&du->dus, &du->agg, du_array);
1087                 duobj = json_array_last_value_object(du_array);
1088                 json_object_add_client_info(duobj, client);
1089         }
1090         if (output_format & FIO_OUTPUT_TERSE)
1091                 print_disk_util(&du->dus, &du->agg, 1, NULL);
1092         if (output_format & FIO_OUTPUT_NORMAL)
1093                 print_disk_util(&du->dus, &du->agg, 0, NULL);
1094 }
1095
1096 static void convert_jobs_eta(struct jobs_eta *je)
1097 {
1098         int i;
1099
1100         je->nr_running          = le32_to_cpu(je->nr_running);
1101         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
1102         je->nr_pending          = le32_to_cpu(je->nr_pending);
1103         je->nr_setting_up       = le32_to_cpu(je->nr_setting_up);
1104         je->files_open          = le32_to_cpu(je->files_open);
1105
1106         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1107                 je->m_rate[i]   = le32_to_cpu(je->m_rate[i]);
1108                 je->t_rate[i]   = le32_to_cpu(je->t_rate[i]);
1109                 je->m_iops[i]   = le32_to_cpu(je->m_iops[i]);
1110                 je->t_iops[i]   = le32_to_cpu(je->t_iops[i]);
1111                 je->rate[i]     = le32_to_cpu(je->rate[i]);
1112                 je->iops[i]     = le32_to_cpu(je->iops[i]);
1113         }
1114
1115         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
1116         je->eta_sec             = le64_to_cpu(je->eta_sec);
1117         je->nr_threads          = le32_to_cpu(je->nr_threads);
1118         je->is_pow2             = le32_to_cpu(je->is_pow2);
1119         je->unit_base           = le32_to_cpu(je->unit_base);
1120 }
1121
1122 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
1123 {
1124         int i;
1125
1126         dst->nr_running         += je->nr_running;
1127         dst->nr_ramp            += je->nr_ramp;
1128         dst->nr_pending         += je->nr_pending;
1129         dst->nr_setting_up      += je->nr_setting_up;
1130         dst->files_open         += je->files_open;
1131
1132         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
1133                 dst->m_rate[i]  += je->m_rate[i];
1134                 dst->t_rate[i]  += je->t_rate[i];
1135                 dst->m_iops[i]  += je->m_iops[i];
1136                 dst->t_iops[i]  += je->t_iops[i];
1137                 dst->rate[i]    += je->rate[i];
1138                 dst->iops[i]    += je->iops[i];
1139         }
1140
1141         dst->elapsed_sec        += je->elapsed_sec;
1142
1143         if (je->eta_sec > dst->eta_sec)
1144                 dst->eta_sec = je->eta_sec;
1145
1146         dst->nr_threads         += je->nr_threads;
1147
1148         /*
1149          * This wont be correct for multiple strings, but at least it
1150          * works for the basic cases.
1151          */
1152         strcpy((char *) dst->run_str, (char *) je->run_str);
1153 }
1154
1155 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
1156 {
1157         struct fio_net_cmd_reply *reply = NULL;
1158         struct flist_head *entry;
1159
1160         flist_for_each(entry, &client->cmd_list) {
1161                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1162
1163                 if (cmd->tag == (uintptr_t) reply)
1164                         break;
1165
1166                 reply = NULL;
1167         }
1168
1169         if (!reply) {
1170                 log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag);
1171                 return;
1172         }
1173
1174         flist_del(&reply->list);
1175         cmd->tag = reply->saved_tag;
1176         free(reply);
1177 }
1178
1179 int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
1180 {
1181         do {
1182                 struct fio_net_cmd_reply *reply = NULL;
1183                 struct flist_head *entry;
1184
1185                 flist_for_each(entry, &client->cmd_list) {
1186                         reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1187
1188                         if (tag == (uintptr_t) reply)
1189                                 break;
1190
1191                         reply = NULL;
1192                 }
1193
1194                 if (!reply)
1195                         break;
1196
1197                 usleep(1000);
1198         } while (1);
1199
1200         return 0;
1201 }
1202
1203 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
1204 {
1205         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1206         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
1207
1208         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
1209
1210         assert(client->eta_in_flight == eta);
1211
1212         client->eta_in_flight = NULL;
1213         flist_del_init(&client->eta_list);
1214         client->eta_timeouts = 0;
1215
1216         if (client->ops->jobs_eta)
1217                 client->ops->jobs_eta(client, je);
1218
1219         fio_client_sum_jobs_eta(&eta->eta, je);
1220         fio_client_dec_jobs_eta(eta, client->ops->eta);
1221 }
1222
1223 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
1224 {
1225         struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload;
1226         const char *os, *arch;
1227         char bit[16];
1228
1229         os = fio_get_os_string(probe->os);
1230         if (!os)
1231                 os = "unknown";
1232
1233         arch = fio_get_arch_string(probe->arch);
1234         if (!arch)
1235                 os = "unknown";
1236
1237         sprintf(bit, "%d-bit", probe->bpp * 8);
1238         probe->flags = le64_to_cpu(probe->flags);
1239
1240         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n",
1241                 probe->hostname, probe->bigendian, bit, os, arch,
1242                 probe->fio_version, (unsigned long) probe->flags);
1243
1244         if (!client->name)
1245                 client->name = strdup((char *) probe->hostname);
1246 }
1247
1248 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
1249 {
1250         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1251
1252         client->state = Client_started;
1253         client->jobs = le32_to_cpu(pdu->jobs);
1254         client->nr_stat = le32_to_cpu(pdu->stat_outputs);
1255
1256         if (client->jobs) {
1257                 int i;
1258
1259                 if (client->opt_lists)
1260                         free(client->opt_lists);
1261
1262                 client->opt_lists = malloc(client->jobs * sizeof(struct flist_head));
1263                 for (i = 0; i < client->jobs; i++)
1264                         INIT_FLIST_HEAD(&client->opt_lists[i]);
1265         }
1266
1267         sum_stat_clients += client->nr_stat;
1268 }
1269
1270 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
1271 {
1272         if (client->error)
1273                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
1274 }
1275
1276 static void convert_stop(struct fio_net_cmd *cmd)
1277 {
1278         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1279
1280         pdu->error = le32_to_cpu(pdu->error);
1281 }
1282
1283 static void convert_text(struct fio_net_cmd *cmd)
1284 {
1285         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
1286
1287         pdu->level      = le32_to_cpu(pdu->level);
1288         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
1289         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
1290         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
1291 }
1292
1293 static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd,
1294                                               struct cmd_iolog_pdu *pdu)
1295 {
1296 #ifdef CONFIG_ZLIB
1297         struct cmd_iolog_pdu *ret;
1298         z_stream stream;
1299         uint32_t nr_samples;
1300         size_t total;
1301         void *p;
1302
1303         stream.zalloc = Z_NULL;
1304         stream.zfree = Z_NULL;
1305         stream.opaque = Z_NULL;
1306         stream.avail_in = 0;
1307         stream.next_in = Z_NULL;
1308
1309         if (inflateInit(&stream) != Z_OK)
1310                 return NULL;
1311
1312         /*
1313          * Get header first, it's not compressed
1314          */
1315         nr_samples = le64_to_cpu(pdu->nr_samples);
1316
1317         total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset));
1318         ret = malloc(total + sizeof(*pdu));
1319         ret->nr_samples = nr_samples;
1320
1321         memcpy(ret, pdu, sizeof(*pdu));
1322
1323         p = (void *) ret + sizeof(*pdu);
1324
1325         stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1326         stream.next_in = (void *) pdu + sizeof(*pdu);
1327         while (stream.avail_in) {
1328                 unsigned int this_chunk = 65536;
1329                 unsigned int this_len;
1330                 int err;
1331
1332                 if (this_chunk > total)
1333                         this_chunk = total;
1334
1335                 stream.avail_out = this_chunk;
1336                 stream.next_out = p;
1337                 err = inflate(&stream, Z_NO_FLUSH);
1338                 /* may be Z_OK, or Z_STREAM_END */
1339                 if (err < 0) {
1340                         log_err("fio: inflate error %d\n", err);
1341                         free(ret);
1342                         ret = NULL;
1343                         goto err;
1344                 }
1345
1346                 this_len = this_chunk - stream.avail_out;
1347                 p += this_len;
1348                 total -= this_len;
1349         }
1350
1351 err:
1352         inflateEnd(&stream);
1353         return ret;
1354 #else
1355         return NULL;
1356 #endif
1357 }
1358
1359 /*
1360  * This has been compressed on the server side, since it can be big.
1361  * Uncompress here.
1362  */
1363 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd)
1364 {
1365         struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1366         struct cmd_iolog_pdu *ret;
1367         uint64_t i;
1368         void *samples;
1369
1370         /*
1371          * Convert if compressed and we support it. If it's not
1372          * compressed, we need not do anything.
1373          */
1374         if (le32_to_cpu(pdu->compressed)) {
1375 #ifndef CONFIG_ZLIB
1376                 log_err("fio: server sent compressed data by mistake\n");
1377                 return NULL;
1378 #endif
1379                 ret = convert_iolog_gz(cmd, pdu);
1380                 if (!ret) {
1381                         log_err("fio: failed decompressing log\n");
1382                         return NULL;
1383                 }
1384         } else
1385                 ret = pdu;
1386
1387         ret->nr_samples         = le64_to_cpu(ret->nr_samples);
1388         ret->thread_number      = le32_to_cpu(ret->thread_number);
1389         ret->log_type           = le32_to_cpu(ret->log_type);
1390         ret->compressed         = le32_to_cpu(ret->compressed);
1391         ret->log_offset         = le32_to_cpu(ret->log_offset);
1392
1393         samples = &ret->samples[0];
1394         for (i = 0; i < ret->nr_samples; i++) {
1395                 struct io_sample *s;
1396
1397                 s = __get_sample(samples, ret->log_offset, i);
1398                 s->time         = le64_to_cpu(s->time);
1399                 s->val          = le64_to_cpu(s->val);
1400                 s->__ddir       = le32_to_cpu(s->__ddir);
1401                 s->bs           = le32_to_cpu(s->bs);
1402
1403                 if (ret->log_offset) {
1404                         struct io_sample_offset *so = (void *) s;
1405
1406                         so->offset = le64_to_cpu(so->offset);
1407                 }
1408         }
1409
1410         return ret;
1411 }
1412
1413 static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep,
1414                            size_t size, uint64_t tag)
1415 {
1416         rep->error = cpu_to_le32(rep->error);
1417         fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL);
1418 }
1419
1420 static int send_file(struct fio_client *client, struct cmd_sendfile *pdu,
1421                      uint64_t tag)
1422 {
1423         struct cmd_sendfile_reply *rep;
1424         struct stat sb;
1425         size_t size;
1426         int fd;
1427
1428         size = sizeof(*rep);
1429         rep = malloc(size);
1430
1431         if (stat((char *)pdu->path, &sb) < 0) {
1432 fail:
1433                 rep->error = errno;
1434                 sendfile_reply(client->fd, rep, size, tag);
1435                 free(rep);
1436                 return 1;
1437         }
1438
1439         size += sb.st_size;
1440         rep = realloc(rep, size);
1441         rep->size = cpu_to_le32((uint32_t) sb.st_size);
1442
1443         fd = open((char *)pdu->path, O_RDONLY);
1444         if (fd == -1 )
1445                 goto fail;
1446
1447         rep->error = read_data(fd, &rep->data, sb.st_size);
1448         sendfile_reply(client->fd, rep, size, tag);
1449         free(rep);
1450         close(fd);
1451         return 0;
1452 }
1453
1454 int fio_handle_client(struct fio_client *client)
1455 {
1456         struct client_ops *ops = client->ops;
1457         struct fio_net_cmd *cmd;
1458
1459         dprint(FD_NET, "client: handle %s\n", client->hostname);
1460
1461         cmd = fio_net_recv_cmd(client->fd);
1462         if (!cmd)
1463                 return 0;
1464
1465         dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1466                 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1467
1468         switch (cmd->opcode) {
1469         case FIO_NET_CMD_QUIT:
1470                 if (ops->quit)
1471                         ops->quit(client, cmd);
1472                 remove_client(client);
1473                 break;
1474         case FIO_NET_CMD_TEXT:
1475                 convert_text(cmd);
1476                 ops->text(client, cmd);
1477                 break;
1478         case FIO_NET_CMD_DU: {
1479                 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1480
1481                 convert_dus(&du->dus);
1482                 convert_agg(&du->agg);
1483
1484                 ops->disk_util(client, cmd);
1485                 break;
1486                 }
1487         case FIO_NET_CMD_TS: {
1488                 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1489
1490                 convert_ts(&p->ts, &p->ts);
1491                 convert_gs(&p->rs, &p->rs);
1492
1493                 ops->thread_status(client, cmd);
1494                 break;
1495                 }
1496         case FIO_NET_CMD_GS: {
1497                 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1498
1499                 convert_gs(gs, gs);
1500
1501                 ops->group_stats(client, cmd);
1502                 break;
1503                 }
1504         case FIO_NET_CMD_ETA: {
1505                 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1506
1507                 remove_reply_cmd(client, cmd);
1508                 convert_jobs_eta(je);
1509                 handle_eta(client, cmd);
1510                 break;
1511                 }
1512         case FIO_NET_CMD_PROBE:
1513                 remove_reply_cmd(client, cmd);
1514                 ops->probe(client, cmd);
1515                 break;
1516         case FIO_NET_CMD_SERVER_START:
1517                 client->state = Client_running;
1518                 if (ops->job_start)
1519                         ops->job_start(client, cmd);
1520                 break;
1521         case FIO_NET_CMD_START: {
1522                 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1523
1524                 pdu->jobs = le32_to_cpu(pdu->jobs);
1525                 ops->start(client, cmd);
1526                 break;
1527                 }
1528         case FIO_NET_CMD_STOP: {
1529                 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1530
1531                 convert_stop(cmd);
1532                 client->state = Client_stopped;
1533                 client->error = le32_to_cpu(pdu->error);
1534                 client->signal = le32_to_cpu(pdu->signal);
1535                 ops->stop(client, cmd);
1536                 break;
1537                 }
1538         case FIO_NET_CMD_ADD_JOB: {
1539                 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1540
1541                 client->thread_number = le32_to_cpu(pdu->thread_number);
1542                 client->groupid = le32_to_cpu(pdu->groupid);
1543
1544                 if (ops->add_job)
1545                         ops->add_job(client, cmd);
1546                 break;
1547                 }
1548         case FIO_NET_CMD_IOLOG:
1549                 if (ops->iolog) {
1550                         struct cmd_iolog_pdu *pdu;
1551
1552                         pdu = convert_iolog(cmd);
1553                         ops->iolog(client, pdu);
1554                 }
1555                 break;
1556         case FIO_NET_CMD_UPDATE_JOB:
1557                 ops->update_job(client, cmd);
1558                 remove_reply_cmd(client, cmd);
1559                 break;
1560         case FIO_NET_CMD_VTRIGGER: {
1561                 struct all_io_list *pdu = (struct all_io_list *) cmd->payload;
1562                 char buf[128];
1563                 int off = 0;
1564
1565                 if (aux_path) {
1566                         strcpy(buf, aux_path);
1567                         off = strlen(buf);
1568                 }
1569
1570                 __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off));
1571                 exec_trigger(trigger_cmd);
1572                 break;
1573                 }
1574         case FIO_NET_CMD_SENDFILE: {
1575                 struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload;
1576                 send_file(client, pdu, cmd->tag);
1577                 break;
1578                 }
1579         case FIO_NET_CMD_JOB_OPT: {
1580                 handle_job_opt(client, cmd);
1581                 break;
1582         }
1583         default:
1584                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1585                 break;
1586         }
1587
1588         free(cmd);
1589         return 1;
1590 }
1591
1592 int fio_clients_send_trigger(const char *cmd)
1593 {
1594         struct flist_head *entry;
1595         struct fio_client *client;
1596         size_t slen;
1597
1598         dprint(FD_NET, "client: send vtrigger: %s\n", cmd);
1599
1600         if (!cmd)
1601                 slen = 0;
1602         else
1603                 slen = strlen(cmd);
1604
1605         flist_for_each(entry, &client_list) {
1606                 struct cmd_vtrigger_pdu *pdu;
1607
1608                 client = flist_entry(entry, struct fio_client, list);
1609
1610                 pdu = malloc(sizeof(*pdu) + slen);
1611                 pdu->len = cpu_to_le16((uint16_t) slen);
1612                 if (slen)
1613                         memcpy(pdu->cmd, cmd, slen);
1614                 fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu,
1615                                         sizeof(*pdu) + slen, NULL, NULL);
1616                 free(pdu);
1617         }
1618
1619         return 0;
1620 }
1621
1622 static void request_client_etas(struct client_ops *ops)
1623 {
1624         struct fio_client *client;
1625         struct flist_head *entry;
1626         struct client_eta *eta;
1627         int skipped = 0;
1628
1629         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1630
1631         eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS));
1632         eta->pending = nr_clients;
1633
1634         flist_for_each(entry, &client_list) {
1635                 client = flist_entry(entry, struct fio_client, list);
1636
1637                 if (!flist_empty(&client->eta_list)) {
1638                         skipped++;
1639                         continue;
1640                 }
1641                 if (client->state != Client_running)
1642                         continue;
1643
1644                 assert(!client->eta_in_flight);
1645                 flist_add_tail(&client->eta_list, &eta_list);
1646                 client->eta_in_flight = eta;
1647                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1648                                         (uintptr_t) eta, &client->cmd_list);
1649         }
1650
1651         while (skipped--) {
1652                 if (!fio_client_dec_jobs_eta(eta, ops->eta))
1653                         break;
1654         }
1655
1656         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1657 }
1658
1659 /*
1660  * A single SEND_ETA timeout isn't fatal. Attempt to recover.
1661  */
1662 static int handle_cmd_timeout(struct fio_client *client,
1663                               struct fio_net_cmd_reply *reply)
1664 {
1665         if (reply->opcode != FIO_NET_CMD_SEND_ETA)
1666                 return 1;
1667
1668         log_info("client <%s>: timeout on SEND_ETA\n", client->hostname);
1669         flist_del(&reply->list);
1670         free(reply);
1671
1672         flist_del_init(&client->eta_list);
1673         if (client->eta_in_flight) {
1674                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
1675                 client->eta_in_flight = NULL;
1676         }
1677
1678         /*
1679          * If we fail 5 in a row, give up...
1680          */
1681         if (client->eta_timeouts++ > 5)
1682                 return 1;
1683
1684         return 0;
1685 }
1686
1687 static int client_check_cmd_timeout(struct fio_client *client,
1688                                     struct timeval *now)
1689 {
1690         struct fio_net_cmd_reply *reply;
1691         struct flist_head *entry, *tmp;
1692         int ret = 0;
1693
1694         flist_for_each_safe(entry, tmp, &client->cmd_list) {
1695                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1696
1697                 if (mtime_since(&reply->tv, now) < FIO_NET_CLIENT_TIMEOUT)
1698                         continue;
1699
1700                 if (!handle_cmd_timeout(client, reply))
1701                         continue;
1702
1703                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1704                                                 fio_server_op(reply->opcode));
1705                 flist_del(&reply->list);
1706                 free(reply);
1707                 ret = 1;
1708         }
1709
1710         return flist_empty(&client->cmd_list) && ret;
1711 }
1712
1713 static int fio_check_clients_timed_out(void)
1714 {
1715         struct fio_client *client;
1716         struct flist_head *entry, *tmp;
1717         struct timeval tv;
1718         int ret = 0;
1719
1720         fio_gettime(&tv, NULL);
1721
1722         flist_for_each_safe(entry, tmp, &client_list) {
1723                 client = flist_entry(entry, struct fio_client, list);
1724
1725                 if (flist_empty(&client->cmd_list))
1726                         continue;
1727
1728                 if (!client_check_cmd_timeout(client, &tv))
1729                         continue;
1730
1731                 if (client->ops->timed_out)
1732                         client->ops->timed_out(client);
1733                 else
1734                         log_err("fio: client %s timed out\n", client->hostname);
1735
1736                 client->error = ETIMEDOUT;
1737                 remove_client(client);
1738                 ret = 1;
1739         }
1740
1741         return ret;
1742 }
1743
1744 int fio_handle_clients(struct client_ops *ops)
1745 {
1746         struct pollfd *pfds;
1747         int i, ret = 0, retval = 0;
1748
1749         fio_gettime(&eta_tv, NULL);
1750
1751         pfds = malloc(nr_clients * sizeof(struct pollfd));
1752
1753         init_thread_stat(&client_ts);
1754         init_group_run_stat(&client_gs);
1755
1756         while (!exit_backend && nr_clients) {
1757                 struct flist_head *entry, *tmp;
1758                 struct fio_client *client;
1759
1760                 i = 0;
1761                 flist_for_each_safe(entry, tmp, &client_list) {
1762                         client = flist_entry(entry, struct fio_client, list);
1763
1764                         if (!client->sent_job && !client->ops->stay_connected &&
1765                             flist_empty(&client->cmd_list)) {
1766                                 remove_client(client);
1767                                 continue;
1768                         }
1769
1770                         pfds[i].fd = client->fd;
1771                         pfds[i].events = POLLIN;
1772                         i++;
1773                 }
1774
1775                 if (!nr_clients)
1776                         break;
1777
1778                 assert(i == nr_clients);
1779
1780                 do {
1781                         struct timeval tv;
1782                         int timeout;
1783
1784                         fio_gettime(&tv, NULL);
1785                         if (mtime_since(&eta_tv, &tv) >= 900) {
1786                                 request_client_etas(ops);
1787                                 memcpy(&eta_tv, &tv, sizeof(tv));
1788
1789                                 if (fio_check_clients_timed_out())
1790                                         break;
1791                         }
1792
1793                         check_trigger_file();
1794
1795                         timeout = min(100u, ops->eta_msec);
1796
1797                         ret = poll(pfds, nr_clients, timeout);
1798                         if (ret < 0) {
1799                                 if (errno == EINTR)
1800                                         continue;
1801                                 log_err("fio: poll clients: %s\n", strerror(errno));
1802                                 break;
1803                         } else if (!ret)
1804                                 continue;
1805                 } while (ret <= 0);
1806
1807                 for (i = 0; i < nr_clients; i++) {
1808                         if (!(pfds[i].revents & POLLIN))
1809                                 continue;
1810
1811                         client = find_client_by_fd(pfds[i].fd);
1812                         if (!client) {
1813                                 log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd);
1814                                 continue;
1815                         }
1816                         if (!fio_handle_client(client)) {
1817                                 log_info("client: host=%s disconnected\n",
1818                                                 client->hostname);
1819                                 remove_client(client);
1820                                 retval = 1;
1821                         } else if (client->error)
1822                                 retval = 1;
1823                         fio_put_client(client);
1824                 }
1825         }
1826
1827         fio_client_json_fini();
1828
1829         free(pfds);
1830         return retval || error_clients;
1831 }