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 #include <zlib.h>
  18
  19 #include "fio.h"
  20 #include "client.h"
  21 #include "server.h"
  22 #include "flist.h"
  23 #include "hash.h"
  24
  25 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
  26 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd);
  27 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd);
  28 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
  29 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd);
  30 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd);
  31 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd);
  32
  33 struct client_ops fio_client_ops = {
  34         .text           = handle_text,
  35         .disk_util      = handle_du,
  36         .thread_status  = handle_ts,
  37         .group_stats    = handle_gs,
  38         .stop           = handle_stop,
  39         .start          = handle_start,
  40         .eta            = display_thread_status,
  41         .probe          = handle_probe,
  42         .eta_msec       = FIO_CLIENT_DEF_ETA_MSEC,
  43         .client_type    = FIO_CLIENT_TYPE_CLI,
  44 };
  45
  46 static struct timeval eta_tv;
  47
  48 static FLIST_HEAD(client_list);
  49 static FLIST_HEAD(eta_list);
  50
  51 static FLIST_HEAD(arg_list);
  52
  53 struct thread_stat client_ts;
  54 struct group_run_stats client_gs;
  55 int sum_stat_clients;
  56
  57 static int sum_stat_nr;
  58
  59 #define FIO_CLIENT_HASH_BITS    7
  60 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  61 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  62 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  63
  64 static void fio_client_add_hash(struct fio_client *client)
  65 {
  66         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  67
  68         bucket &= FIO_CLIENT_HASH_MASK;
  69         flist_add(&client->hash_list, &client_hash[bucket]);
  70 }
  71
  72 static void fio_client_remove_hash(struct fio_client *client)
  73 {
  74         if (!flist_empty(&client->hash_list))
  75                 flist_del_init(&client->hash_list);
  76 }
  77
  78 static void fio_init fio_client_hash_init(void)
  79 {
  80         int i;
  81
  82         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
  83                 INIT_FLIST_HEAD(&client_hash[i]);
  84 }
  85
  86 static struct fio_client *find_client_by_fd(int fd)
  87 {
  88         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
  89         struct fio_client *client;
  90         struct flist_head *entry;
  91
  92         flist_for_each(entry, &client_hash[bucket]) {
  93                 client = flist_entry(entry, struct fio_client, hash_list);
  94
  95                 if (client->fd == fd) {
  96                         client->refs++;
  97                         return client;
  98                 }
  99         }
 100
 101         return NULL;
 102 }
 103
 104 void fio_put_client(struct fio_client *client)
 105 {
 106         if (--client->refs)
 107                 return;
 108
 109         free(client->hostname);
 110         if (client->argv)
 111                 free(client->argv);
 112         if (client->name)
 113                 free(client->name);
 114
 115         free(client);
 116 }
 117
 118 static void remove_client(struct fio_client *client)
 119 {
 120         assert(client->refs);
 121
 122         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 123
 124         if (!flist_empty(&client->list))
 125                 flist_del_init(&client->list);
 126
 127         fio_client_remove_hash(client);
 128
 129         if (!flist_empty(&client->eta_list)) {
 130                 flist_del_init(&client->eta_list);
 131                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
 132         }
 133
 134         nr_clients--;
 135         sum_stat_clients--;
 136
 137         fio_put_client(client);
 138 }
 139
 140 struct fio_client *fio_get_client(struct fio_client *client)
 141 {
 142         client->refs++;
 143         return client;
 144 }
 145
 146 static void __fio_client_add_cmd_option(struct fio_client *client,
 147                                         const char *opt)
 148 {
 149         int index;
 150
 151         index = client->argc++;
 152         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 153         client->argv[index] = strdup(opt);
 154         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 155 }
 156
 157 void fio_client_add_cmd_option(void *cookie, const char *opt)
 158 {
 159         struct fio_client *client = cookie;
 160         struct flist_head *entry;
 161
 162         if (!client || !opt)
 163                 return;
 164
 165         __fio_client_add_cmd_option(client, opt);
 166
 167         /*
 168          * Duplicate arguments to shared client group
 169          */
 170         flist_for_each(entry, &arg_list) {
 171                 client = flist_entry(entry, struct fio_client, arg_list);
 172
 173                 __fio_client_add_cmd_option(client, opt);
 174         }
 175 }
 176
 177 struct fio_client *fio_client_add_explicit(struct client_ops *ops,
 178                                            const char *hostname, int type,
 179                                            int port)
 180 {
 181         struct fio_client *client;
 182
 183         client = malloc(sizeof(*client));
 184         memset(client, 0, sizeof(*client));
 185
 186         INIT_FLIST_HEAD(&client->list);
 187         INIT_FLIST_HEAD(&client->hash_list);
 188         INIT_FLIST_HEAD(&client->arg_list);
 189         INIT_FLIST_HEAD(&client->eta_list);
 190         INIT_FLIST_HEAD(&client->cmd_list);
 191
 192         client->hostname = strdup(hostname);
 193
 194         if (type == Fio_client_socket)
 195                 client->is_sock = 1;
 196         else {
 197                 int ipv6;
 198
 199                 ipv6 = type == Fio_client_ipv6;
 200                 if (fio_server_parse_host(hostname, &ipv6,
 201                                                 &client->addr.sin_addr,
 202                                                 &client->addr6.sin6_addr))
 203                         goto err;
 204
 205                 client->port = port;
 206         }
 207
 208         client->fd = -1;
 209         client->ops = ops;
 210         client->refs = 1;
 211         client->type = ops->client_type;
 212
 213         __fio_client_add_cmd_option(client, "fio");
 214
 215         flist_add(&client->list, &client_list);
 216         nr_clients++;
 217         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 218         return client;
 219 err:
 220         free(client);
 221         return NULL;
 222 }
 223
 224 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
 225 {
 226         struct fio_client *existing = *cookie;
 227         struct fio_client *client;
 228
 229         if (existing) {
 230                 /*
 231                  * We always add our "exec" name as the option, hence 1
 232                  * means empty.
 233                  */
 234                 if (existing->argc == 1)
 235                         flist_add_tail(&existing->arg_list, &arg_list);
 236                 else {
 237                         while (!flist_empty(&arg_list))
 238                                 flist_del_init(arg_list.next);
 239                 }
 240         }
 241
 242         client = malloc(sizeof(*client));
 243         memset(client, 0, sizeof(*client));
 244
 245         INIT_FLIST_HEAD(&client->list);
 246         INIT_FLIST_HEAD(&client->hash_list);
 247         INIT_FLIST_HEAD(&client->arg_list);
 248         INIT_FLIST_HEAD(&client->eta_list);
 249         INIT_FLIST_HEAD(&client->cmd_list);
 250
 251         if (fio_server_parse_string(hostname, &client->hostname,
 252                                         &client->is_sock, &client->port,
 253                                         &client->addr.sin_addr,
 254                                         &client->addr6.sin6_addr,
 255                                         &client->ipv6))
 256                 return -1;
 257
 258         client->fd = -1;
 259         client->ops = ops;
 260         client->refs = 1;
 261         client->type = ops->client_type;
 262
 263         __fio_client_add_cmd_option(client, "fio");
 264
 265         flist_add(&client->list, &client_list);
 266         nr_clients++;
 267         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 268         *cookie = client;
 269         return 0;
 270 }
 271
 272 static void probe_client(struct fio_client *client)
 273 {
 274         dprint(FD_NET, "client: send probe\n");
 275
 276         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
 277 }
 278
 279 static int fio_client_connect_ip(struct fio_client *client)
 280 {
 281         struct sockaddr *addr;
 282         fio_socklen_t socklen;
 283         int fd, domain;
 284
 285         if (client->ipv6) {
 286                 client->addr6.sin6_family = AF_INET6;
 287                 client->addr6.sin6_port = htons(client->port);
 288                 domain = AF_INET6;
 289                 addr = (struct sockaddr *) &client->addr6;
 290                 socklen = sizeof(client->addr6);
 291         } else {
 292                 client->addr.sin_family = AF_INET;
 293                 client->addr.sin_port = htons(client->port);
 294                 domain = AF_INET;
 295                 addr = (struct sockaddr *) &client->addr;
 296                 socklen = sizeof(client->addr);
 297         }
 298
 299         fd = socket(domain, SOCK_STREAM, 0);
 300         if (fd < 0) {
 301                 int ret = -errno;
 302
 303                 log_err("fio: socket: %s\n", strerror(errno));
 304                 return ret;
 305         }
 306
 307         if (connect(fd, addr, socklen) < 0) {
 308                 int ret = -errno;
 309
 310                 log_err("fio: connect: %s\n", strerror(errno));
 311                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 312                                                                 client->port);
 313                 close(fd);
 314                 return ret;
 315         }
 316
 317         return fd;
 318 }
 319
 320 static int fio_client_connect_sock(struct fio_client *client)
 321 {
 322         struct sockaddr_un *addr = &client->addr_un;
 323         fio_socklen_t len;
 324         int fd;
 325
 326         memset(addr, 0, sizeof(*addr));
 327         addr->sun_family = AF_UNIX;
 328         strcpy(addr->sun_path, client->hostname);
 329
 330         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 331         if (fd < 0) {
 332                 int ret = -errno;
 333
 334                 log_err("fio: socket: %s\n", strerror(errno));
 335                 return ret;
 336         }
 337
 338         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 339         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 340                 int ret = -errno;
 341
 342                 log_err("fio: connect; %s\n", strerror(errno));
 343                 close(fd);
 344                 return ret;
 345         }
 346
 347         return fd;
 348 }
 349
 350 int fio_client_connect(struct fio_client *client)
 351 {
 352         int fd;
 353
 354         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 355
 356         if (client->is_sock)
 357                 fd = fio_client_connect_sock(client);
 358         else
 359                 fd = fio_client_connect_ip(client);
 360
 361         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 362
 363         if (fd < 0)
 364                 return fd;
 365
 366         client->fd = fd;
 367         fio_client_add_hash(client);
 368         client->state = Client_connected;
 369
 370         probe_client(client);
 371         return 0;
 372 }
 373
 374 void fio_client_terminate(struct fio_client *client)
 375 {
 376         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0, NULL);
 377 }
 378
 379 void fio_clients_terminate(void)
 380 {
 381         struct flist_head *entry;
 382         struct fio_client *client;
 383
 384         dprint(FD_NET, "client: terminate clients\n");
 385
 386         flist_for_each(entry, &client_list) {
 387                 client = flist_entry(entry, struct fio_client, list);
 388                 fio_client_terminate(client);
 389         }
 390 }
 391
 392 static void sig_int(int sig)
 393 {
 394         dprint(FD_NET, "client: got signal %d\n", sig);
 395         fio_clients_terminate();
 396 }
 397
 398 static void client_signal_handler(void)
 399 {
 400         struct sigaction act;
 401
 402         memset(&act, 0, sizeof(act));
 403         act.sa_handler = sig_int;
 404         act.sa_flags = SA_RESTART;
 405         sigaction(SIGINT, &act, NULL);
 406
 407         memset(&act, 0, sizeof(act));
 408         act.sa_handler = sig_int;
 409         act.sa_flags = SA_RESTART;
 410         sigaction(SIGTERM, &act, NULL);
 411 }
 412
 413 static int send_client_cmd_line(struct fio_client *client)
 414 {
 415         struct cmd_single_line_pdu *cslp;
 416         struct cmd_line_pdu *clp;
 417         unsigned long offset;
 418         unsigned int *lens;
 419         void *pdu;
 420         size_t mem;
 421         int i, ret;
 422
 423         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 424
 425         lens = malloc(client->argc * sizeof(unsigned int));
 426
 427         /*
 428          * Find out how much mem we need
 429          */
 430         for (i = 0, mem = 0; i < client->argc; i++) {
 431                 lens[i] = strlen(client->argv[i]) + 1;
 432                 mem += lens[i];
 433         }
 434
 435         /*
 436          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 437          */
 438         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 439
 440         pdu = malloc(mem);
 441         clp = pdu;
 442         offset = sizeof(*clp);
 443
 444         for (i = 0; i < client->argc; i++) {
 445                 uint16_t arg_len = lens[i];
 446
 447                 cslp = pdu + offset;
 448                 strcpy((char *) cslp->text, client->argv[i]);
 449                 cslp->len = cpu_to_le16(arg_len);
 450                 offset += sizeof(*cslp) + arg_len;
 451         }
 452
 453         free(lens);
 454         clp->lines = cpu_to_le16(client->argc);
 455         clp->client_type = __cpu_to_le16(client->type);
 456         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
 457         free(pdu);
 458         return ret;
 459 }
 460
 461 int fio_clients_connect(void)
 462 {
 463         struct fio_client *client;
 464         struct flist_head *entry, *tmp;
 465         int ret;
 466
 467 #ifdef WIN32
 468         WSADATA wsd;
 469         WSAStartup(MAKEWORD(2,2), &wsd);
 470 #endif
 471
 472         dprint(FD_NET, "client: connect all\n");
 473
 474         client_signal_handler();
 475
 476         flist_for_each_safe(entry, tmp, &client_list) {
 477                 client = flist_entry(entry, struct fio_client, list);
 478
 479                 ret = fio_client_connect(client);
 480                 if (ret) {
 481                         remove_client(client);
 482                         continue;
 483                 }
 484
 485                 if (client->argc > 1)
 486                         send_client_cmd_line(client);
 487         }
 488
 489         return !nr_clients;
 490 }
 491
 492 int fio_start_client(struct fio_client *client)
 493 {
 494         dprint(FD_NET, "client: start %s\n", client->hostname);
 495         return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
 496 }
 497
 498 int fio_start_all_clients(void)
 499 {
 500         struct fio_client *client;
 501         struct flist_head *entry, *tmp;
 502         int ret;
 503
 504         dprint(FD_NET, "client: start all\n");
 505
 506         flist_for_each_safe(entry, tmp, &client_list) {
 507                 client = flist_entry(entry, struct fio_client, list);
 508
 509                 ret = fio_start_client(client);
 510                 if (ret) {
 511                         remove_client(client);
 512                         continue;
 513                 }
 514         }
 515
 516         return flist_empty(&client_list);
 517 }
 518
 519 /*
 520  * Send file contents to server backend. We could use sendfile(), but to remain
 521  * more portable lets just read/write the darn thing.
 522  */
 523 static int __fio_client_send_ini(struct fio_client *client, const char *filename)
 524 {
 525         struct cmd_job_pdu *pdu;
 526         size_t p_size;
 527         struct stat sb;
 528         char *p;
 529         void *buf;
 530         off_t len;
 531         int fd, ret;
 532
 533         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 534
 535         fd = open(filename, O_RDONLY);
 536         if (fd < 0) {
 537                 int ret = -errno;
 538
 539                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 540                 return ret;
 541         }
 542
 543         if (fstat(fd, &sb) < 0) {
 544                 int ret = -errno;
 545
 546                 log_err("fio: job file stat: %s\n", strerror(errno));
 547                 close(fd);
 548                 return ret;
 549         }
 550
 551         p_size = sb.st_size + sizeof(*pdu);
 552         pdu = malloc(p_size);
 553         buf = pdu->buf;
 554
 555         len = sb.st_size;
 556         p = buf;
 557         do {
 558                 ret = read(fd, p, len);
 559                 if (ret > 0) {
 560                         len -= ret;
 561                         if (!len)
 562                                 break;
 563                         p += ret;
 564                         continue;
 565                 } else if (!ret)
 566                         break;
 567                 else if (errno == EAGAIN || errno == EINTR)
 568                         continue;
 569         } while (1);
 570
 571         if (len) {
 572                 log_err("fio: failed reading job file %s\n", filename);
 573                 close(fd);
 574                 free(buf);
 575                 return 1;
 576         }
 577
 578         pdu->buf_len = __cpu_to_le32(sb.st_size);
 579         pdu->client_type = cpu_to_le32(client->type);
 580
 581         client->sent_job = 1;
 582         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, 0);
 583         free(pdu);
 584         close(fd);
 585         return ret;
 586 }
 587
 588 int fio_client_send_ini(struct fio_client *client, const char *filename)
 589 {
 590         int ret;
 591
 592         ret = __fio_client_send_ini(client, filename);
 593         if (!ret)
 594                 client->sent_job = 1;
 595
 596         return ret;
 597 }
 598
 599 int fio_clients_send_ini(const char *filename)
 600 {
 601         struct fio_client *client;
 602         struct flist_head *entry, *tmp;
 603
 604         flist_for_each_safe(entry, tmp, &client_list) {
 605                 client = flist_entry(entry, struct fio_client, list);
 606
 607                 if (fio_client_send_ini(client, filename))
 608                         remove_client(client);
 609         }
 610
 611         return !nr_clients;
 612 }
 613
 614 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 615 {
 616         dst->max_val    = le64_to_cpu(src->max_val);
 617         dst->min_val    = le64_to_cpu(src->min_val);
 618         dst->samples    = le64_to_cpu(src->samples);
 619
 620         /*
 621          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 622          */
 623         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 624         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 625 }
 626
 627 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 628 {
 629         int i, j;
 630
 631         dst->error              = le32_to_cpu(src->error);
 632         dst->thread_number      = le32_to_cpu(src->thread_number);
 633         dst->groupid            = le32_to_cpu(src->groupid);
 634         dst->pid                = le32_to_cpu(src->pid);
 635         dst->members            = le32_to_cpu(src->members);
 636
 637         for (i = 0; i < 2; i++) {
 638                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 639                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 640                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 641                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 642         }
 643
 644         dst->usr_time           = le64_to_cpu(src->usr_time);
 645         dst->sys_time           = le64_to_cpu(src->sys_time);
 646         dst->ctx                = le64_to_cpu(src->ctx);
 647         dst->minf               = le64_to_cpu(src->minf);
 648         dst->majf               = le64_to_cpu(src->majf);
 649         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 650
 651         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 652                 fio_fp64_t *fps = &src->percentile_list[i];
 653                 fio_fp64_t *fpd = &dst->percentile_list[i];
 654
 655                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 656         }
 657
 658         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 659                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 660                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 661                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 662         }
 663
 664         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 665                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 666                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 667         }
 668
 669         for (i = 0; i < 2; i++)
 670                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 671                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 672
 673         for (i = 0; i < 3; i++) {
 674                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 675                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 676         }
 677
 678         dst->total_submit       = le64_to_cpu(src->total_submit);
 679         dst->total_complete     = le64_to_cpu(src->total_complete);
 680
 681         for (i = 0; i < 2; i++) {
 682                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 683                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 684         }
 685
 686         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 687         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 688         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 689         dst->first_error        = le32_to_cpu(src->first_error);
 690         dst->kb_base            = le32_to_cpu(src->kb_base);
 691 }
 692
 693 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 694 {
 695         int i;
 696
 697         for (i = 0; i < 2; i++) {
 698                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 699                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 700                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 701                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 702                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 703                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 704         }
 705
 706         dst->kb_base    = le32_to_cpu(src->kb_base);
 707         dst->groupid    = le32_to_cpu(src->groupid);
 708 }
 709
 710 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
 711 {
 712         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 713
 714         show_thread_status(&p->ts, &p->rs);
 715
 716         if (sum_stat_clients == 1)
 717                 return;
 718
 719         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 720         sum_group_stats(&client_gs, &p->rs);
 721
 722         client_ts.members++;
 723         client_ts.thread_number = p->ts.thread_number;
 724         client_ts.groupid = p->ts.groupid;
 725
 726         if (++sum_stat_nr == sum_stat_clients) {
 727                 strcpy(client_ts.name, "All clients");
 728                 show_thread_status(&client_ts, &client_gs);
 729         }
 730 }
 731
 732 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
 733 {
 734         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 735
 736         show_group_stats(gs);
 737 }
 738
 739 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
 740 {
 741         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
 742         const char *buf = (const char *) pdu->buf;
 743         const char *name;
 744         int fio_unused ret;
 745
 746         name = client->name ? client->name : client->hostname;
 747
 748         if (!client->skip_newline)
 749                 fprintf(f_out, "<%s> ", name);
 750         ret = fwrite(buf, pdu->buf_len, 1, f_out);
 751         fflush(f_out);
 752         client->skip_newline = strchr(buf, '\n') == NULL;
 753 }
 754
 755 static void convert_agg(struct disk_util_agg *agg)
 756 {
 757         int i;
 758
 759         for (i = 0; i < 2; i++) {
 760                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 761                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 762                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 763                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 764         }
 765
 766         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 767         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 768         agg->slavecount         = le32_to_cpu(agg->slavecount);
 769         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 770 }
 771
 772 static void convert_dus(struct disk_util_stat *dus)
 773 {
 774         int i;
 775
 776         for (i = 0; i < 2; i++) {
 777                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
 778                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
 779                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
 780                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
 781         }
 782
 783         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
 784         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
 785         dus->msec               = le64_to_cpu(dus->msec);
 786 }
 787
 788 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 789 {
 790         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 791
 792         if (!client->disk_stats_shown) {
 793                 client->disk_stats_shown = 1;
 794                 log_info("\nDisk stats (read/write):\n");
 795         }
 796
 797         print_disk_util(&du->dus, &du->agg, terse_output);
 798 }
 799
 800 static void convert_jobs_eta(struct jobs_eta *je)
 801 {
 802         int i;
 803
 804         je->nr_running          = le32_to_cpu(je->nr_running);
 805         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 806         je->nr_pending          = le32_to_cpu(je->nr_pending);
 807         je->files_open          = le32_to_cpu(je->files_open);
 808
 809         for (i = 0; i < 2; i++) {
 810                 je->m_rate[i]           = le32_to_cpu(je->m_rate[i]);
 811                 je->t_rate[i]           = le32_to_cpu(je->t_rate[i]);
 812                 je->m_iops[i]           = le32_to_cpu(je->m_iops[i]);
 813                 je->t_iops[i]           = le32_to_cpu(je->t_iops[i]);
 814                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 815                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 816         }
 817
 818         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 819         je->eta_sec             = le64_to_cpu(je->eta_sec);
 820         je->nr_threads          = le32_to_cpu(je->nr_threads);
 821 }
 822
 823 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 824 {
 825         int i;
 826
 827         dst->nr_running         += je->nr_running;
 828         dst->nr_ramp            += je->nr_ramp;
 829         dst->nr_pending         += je->nr_pending;
 830         dst->files_open         += je->files_open;
 831
 832         for (i = 0; i < 2; i++) {
 833                 dst->m_rate[i]  += je->m_rate[i];
 834                 dst->t_rate[i]  += je->t_rate[i];
 835                 dst->m_iops[i]  += je->m_iops[i];
 836                 dst->t_iops[i]  += je->t_iops[i];
 837                 dst->rate[i]    += je->rate[i];
 838                 dst->iops[i]    += je->iops[i];
 839         }
 840
 841         dst->elapsed_sec        += je->elapsed_sec;
 842
 843         if (je->eta_sec > dst->eta_sec)
 844                 dst->eta_sec = je->eta_sec;
 845
 846         dst->nr_threads         += je->nr_threads;
 847         /* we need to handle je->run_str too ... */
 848 }
 849
 850 void fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
 851 {
 852         if (!--eta->pending) {
 853                 eta_fn(&eta->eta);
 854                 free(eta);
 855         }
 856 }
 857
 858 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 859 {
 860         struct fio_net_int_cmd *icmd = NULL;
 861         struct flist_head *entry;
 862
 863         flist_for_each(entry, &client->cmd_list) {
 864                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
 865
 866                 if (cmd->tag == (uintptr_t) icmd)
 867                         break;
 868
 869                 icmd = NULL;
 870         }
 871
 872         if (!icmd) {
 873                 log_err("fio: client: unable to find matching tag\n");
 874                 return;
 875         }
 876
 877         flist_del(&icmd->list);
 878         cmd->tag = icmd->saved_tag;
 879         free(icmd);
 880 }
 881
 882 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 883 {
 884         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 885         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 886
 887         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 888
 889         assert(client->eta_in_flight == eta);
 890
 891         client->eta_in_flight = NULL;
 892         flist_del_init(&client->eta_list);
 893
 894         if (client->ops->jobs_eta)
 895                 client->ops->jobs_eta(client, je);
 896
 897         fio_client_sum_jobs_eta(&eta->eta, je);
 898         fio_client_dec_jobs_eta(eta, client->ops->eta);
 899 }
 900
 901 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 902 {
 903         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 904         const char *os, *arch;
 905         char bit[16];
 906
 907         os = fio_get_os_string(probe->os);
 908         if (!os)
 909                 os = "unknown";
 910
 911         arch = fio_get_arch_string(probe->arch);
 912         if (!arch)
 913                 os = "unknown";
 914
 915         sprintf(bit, "%d-bit", probe->bpp * 8);
 916
 917         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
 918                 probe->hostname, probe->bigendian, bit, os, arch,
 919                 probe->fio_major, probe->fio_minor, probe->fio_patch);
 920
 921         if (!client->name)
 922                 client->name = strdup((char *) probe->hostname);
 923 }
 924
 925 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
 926 {
 927         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 928
 929         client->state = Client_started;
 930         client->jobs = pdu->jobs;
 931 }
 932
 933 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
 934 {
 935         if (client->error)
 936                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
 937 }
 938
 939 static void convert_stop(struct fio_net_cmd *cmd)
 940 {
 941         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
 942
 943         pdu->error = le32_to_cpu(pdu->error);
 944 }
 945
 946 static void convert_text(struct fio_net_cmd *cmd)
 947 {
 948         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
 949
 950         pdu->level      = le32_to_cpu(pdu->level);
 951         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
 952         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
 953         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
 954 }
 955
 956 /*
 957  * This has been compressed on the server side, since it can be big.
 958  * Uncompress here.
 959  */
 960 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd)
 961 {
 962         struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
 963         struct cmd_iolog_pdu *ret;
 964         uint32_t nr_samples;
 965         unsigned long total;
 966         z_stream stream;
 967         void *p;
 968         int i;
 969
 970         stream.zalloc = Z_NULL;
 971         stream.zfree = Z_NULL;
 972         stream.opaque = Z_NULL;
 973         stream.avail_in = 0;
 974         stream.next_in = Z_NULL;
 975
 976         if (inflateInit(&stream) != Z_OK)
 977                 return NULL;
 978
 979         /*
 980          * Get header first, it's not compressed
 981          */
 982         nr_samples = le32_to_cpu(pdu->nr_samples);
 983
 984         total = nr_samples * sizeof(struct io_sample);
 985         ret = malloc(total + sizeof(*pdu));
 986         ret->thread_number = le32_to_cpu(pdu->thread_number);
 987         ret->nr_samples = nr_samples;
 988         ret->log_type = le32_to_cpu(pdu->log_type);
 989         strcpy((char *) ret->name, (char *) pdu->name);
 990
 991         p = (void *) ret + sizeof(*pdu);
 992
 993         stream.avail_in = cmd->pdu_len - sizeof(*pdu);
 994         stream.next_in = (void *) pdu + sizeof(*pdu);
 995         while (stream.avail_in) {
 996                 unsigned int this_chunk = 65536;
 997                 unsigned int this_len;
 998                 int err;
 999
1000                 if (this_chunk > total)
1001                         this_chunk = total;
1002
1003                 stream.avail_out = this_chunk;
1004                 stream.next_out = p;
1005                 err = inflate(&stream, Z_NO_FLUSH);
1006                 /* may be Z_OK, or Z_STREAM_END */
1007                 if (err < 0) {
1008                         log_err("fio: inflate error %d\n", err);
1009                         free(ret);
1010                         ret = NULL;
1011                         goto out;
1012                 }
1013
1014                 this_len = this_chunk - stream.avail_out;
1015                 p += this_len;
1016                 total -= this_len;
1017         }
1018
1019         for (i = 0; i < ret->nr_samples; i++) {
1020                 struct io_sample *s = &ret->samples[i];
1021
1022                 s->time = le64_to_cpu(s->time);
1023                 s->val  = le64_to_cpu(s->val);
1024                 s->ddir = le32_to_cpu(s->ddir);
1025                 s->bs   = le32_to_cpu(s->bs);
1026         }
1027
1028 out:
1029         inflateEnd(&stream);
1030         return ret;
1031 }
1032
1033 int fio_handle_client(struct fio_client *client)
1034 {
1035         struct client_ops *ops = client->ops;
1036         struct fio_net_cmd *cmd;
1037
1038         dprint(FD_NET, "client: handle %s\n", client->hostname);
1039
1040         cmd = fio_net_recv_cmd(client->fd);
1041         if (!cmd)
1042                 return 0;
1043
1044         dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1045                 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1046
1047         switch (cmd->opcode) {
1048         case FIO_NET_CMD_QUIT:
1049                 if (ops->quit)
1050                         ops->quit(client, cmd);
1051                 remove_client(client);
1052                 free(cmd);
1053                 break;
1054         case FIO_NET_CMD_TEXT:
1055                 convert_text(cmd);
1056                 ops->text(client, cmd);
1057                 free(cmd);
1058                 break;
1059         case FIO_NET_CMD_DU: {
1060                 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1061
1062                 convert_dus(&du->dus);
1063                 convert_agg(&du->agg);
1064
1065                 ops->disk_util(client, cmd);
1066                 free(cmd);
1067                 break;
1068                 }
1069         case FIO_NET_CMD_TS: {
1070                 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1071
1072                 convert_ts(&p->ts, &p->ts);
1073                 convert_gs(&p->rs, &p->rs);
1074
1075                 ops->thread_status(client, cmd);
1076                 free(cmd);
1077                 break;
1078                 }
1079         case FIO_NET_CMD_GS: {
1080                 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1081
1082                 convert_gs(gs, gs);
1083
1084                 ops->group_stats(client, cmd);
1085                 free(cmd);
1086                 break;
1087                 }
1088         case FIO_NET_CMD_ETA: {
1089                 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1090
1091                 remove_reply_cmd(client, cmd);
1092                 convert_jobs_eta(je);
1093                 handle_eta(client, cmd);
1094                 free(cmd);
1095                 break;
1096                 }
1097         case FIO_NET_CMD_PROBE:
1098                 remove_reply_cmd(client, cmd);
1099                 ops->probe(client, cmd);
1100                 free(cmd);
1101                 break;
1102         case FIO_NET_CMD_SERVER_START:
1103                 client->state = Client_running;
1104                 if (ops->job_start)
1105                         ops->job_start(client, cmd);
1106                 free(cmd);
1107                 break;
1108         case FIO_NET_CMD_START: {
1109                 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1110
1111                 pdu->jobs = le32_to_cpu(pdu->jobs);
1112                 ops->start(client, cmd);
1113                 free(cmd);
1114                 break;
1115                 }
1116         case FIO_NET_CMD_STOP: {
1117                 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1118
1119                 convert_stop(cmd);
1120                 client->state = Client_stopped;
1121                 client->error = pdu->error;
1122                 ops->stop(client, cmd);
1123                 free(cmd);
1124                 break;
1125                 }
1126         case FIO_NET_CMD_ADD_JOB:
1127                 if (ops->add_job)
1128                         ops->add_job(client, cmd);
1129                 free(cmd);
1130                 break;
1131         case FIO_NET_CMD_IOLOG:
1132                 if (ops->iolog) {
1133                         struct cmd_iolog_pdu *pdu;
1134
1135                         pdu = convert_iolog(cmd);
1136                         ops->iolog(client, pdu);
1137                 }
1138                 free(cmd);
1139                 break;
1140         default:
1141                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1142                 free(cmd);
1143                 break;
1144         }
1145
1146         return 1;
1147 }
1148
1149 static void request_client_etas(struct client_ops *ops)
1150 {
1151         struct fio_client *client;
1152         struct flist_head *entry;
1153         struct client_eta *eta;
1154         int skipped = 0;
1155
1156         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1157
1158         eta = malloc(sizeof(*eta));
1159         memset(&eta->eta, 0, sizeof(eta->eta));
1160         eta->pending = nr_clients;
1161
1162         flist_for_each(entry, &client_list) {
1163                 client = flist_entry(entry, struct fio_client, list);
1164
1165                 if (!flist_empty(&client->eta_list)) {
1166                         skipped++;
1167                         continue;
1168                 }
1169                 if (client->state != Client_running)
1170                         continue;
1171
1172                 assert(!client->eta_in_flight);
1173                 flist_add_tail(&client->eta_list, &eta_list);
1174                 client->eta_in_flight = eta;
1175                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1176                                         (uintptr_t) eta, &client->cmd_list);
1177         }
1178
1179         while (skipped--)
1180                 fio_client_dec_jobs_eta(eta, ops->eta);
1181
1182         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1183 }
1184
1185 static int client_check_cmd_timeout(struct fio_client *client,
1186                                     struct timeval *now)
1187 {
1188         struct fio_net_int_cmd *cmd;
1189         struct flist_head *entry, *tmp;
1190         int ret = 0;
1191
1192         flist_for_each_safe(entry, tmp, &client->cmd_list) {
1193                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
1194
1195                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
1196                         continue;
1197
1198                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1199                                                 fio_server_op(cmd->cmd.opcode));
1200                 flist_del(&cmd->list);
1201                 free(cmd);
1202                 ret = 1;
1203         }
1204
1205         return flist_empty(&client->cmd_list) && ret;
1206 }
1207
1208 static int fio_check_clients_timed_out(void)
1209 {
1210         struct fio_client *client;
1211         struct flist_head *entry, *tmp;
1212         struct timeval tv;
1213         int ret = 0;
1214
1215         gettimeofday(&tv, NULL);
1216
1217         flist_for_each_safe(entry, tmp, &client_list) {
1218                 client = flist_entry(entry, struct fio_client, list);
1219
1220                 if (flist_empty(&client->cmd_list))
1221                         continue;
1222
1223                 if (!client_check_cmd_timeout(client, &tv))
1224                         continue;
1225
1226                 if (client->ops->timed_out)
1227                         client->ops->timed_out(client);
1228                 else
1229                         log_err("fio: client %s timed out\n", client->hostname);
1230
1231                 remove_client(client);
1232                 ret = 1;
1233         }
1234
1235         return ret;
1236 }
1237
1238 int fio_handle_clients(struct client_ops *ops)
1239 {
1240         struct pollfd *pfds;
1241         int i, ret = 0, retval = 0;
1242
1243         gettimeofday(&eta_tv, NULL);
1244
1245         pfds = malloc(nr_clients * sizeof(struct pollfd));
1246
1247         sum_stat_clients = nr_clients;
1248         init_thread_stat(&client_ts);
1249         init_group_run_stat(&client_gs);
1250
1251         while (!exit_backend && nr_clients) {
1252                 struct flist_head *entry, *tmp;
1253                 struct fio_client *client;
1254
1255                 i = 0;
1256                 flist_for_each_safe(entry, tmp, &client_list) {
1257                         client = flist_entry(entry, struct fio_client, list);
1258
1259                         if (!client->sent_job && !client->ops->stay_connected &&
1260                             flist_empty(&client->cmd_list)) {
1261                                 remove_client(client);
1262                                 continue;
1263                         }
1264
1265                         pfds[i].fd = client->fd;
1266                         pfds[i].events = POLLIN;
1267                         i++;
1268                 }
1269
1270                 if (!nr_clients)
1271                         break;
1272
1273                 assert(i == nr_clients);
1274
1275                 do {
1276                         struct timeval tv;
1277
1278                         gettimeofday(&tv, NULL);
1279                         if (mtime_since(&eta_tv, &tv) >= ops->eta_msec) {
1280                                 request_client_etas(ops);
1281                                 memcpy(&eta_tv, &tv, sizeof(tv));
1282
1283                                 if (fio_check_clients_timed_out())
1284                                         break;
1285                         }
1286
1287                         ret = poll(pfds, nr_clients, 100);
1288                         if (ret < 0) {
1289                                 if (errno == EINTR)
1290                                         continue;
1291                                 log_err("fio: poll clients: %s\n", strerror(errno));
1292                                 break;
1293                         } else if (!ret)
1294                                 continue;
1295                 } while (ret <= 0);
1296
1297                 for (i = 0; i < nr_clients; i++) {
1298                         if (!(pfds[i].revents & POLLIN))
1299                                 continue;
1300
1301                         client = find_client_by_fd(pfds[i].fd);
1302                         if (!client) {
1303                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1304                                 continue;
1305                         }
1306                         if (!fio_handle_client(client)) {
1307                                 log_info("client: host=%s disconnected\n",
1308                                                 client->hostname);
1309                                 remove_client(client);
1310                                 retval = 1;
1311                         } else if (client->error)
1312                                 retval = 1;
1313                         fio_put_client(client);
1314                 }
1315         }
1316
1317         free(pfds);
1318         return retval;
1319 }