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
  18 #include "fio.h"
  19 #include "client.h"
  20 #include "server.h"
  21 #include "flist.h"
  22 #include "hash.h"
  23
  24 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
  25 static void handle_ts(struct fio_net_cmd *cmd);
  26 static void handle_gs(struct fio_net_cmd *cmd);
  27 static void handle_eta(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
  31 struct client_ops fio_client_ops = {
  32         .text_op        = handle_text,
  33         .disk_util      = handle_du,
  34         .thread_status  = handle_ts,
  35         .group_stats    = handle_gs,
  36         .eta            = handle_eta,
  37         .probe          = handle_probe,
  38 };
  39
  40 static struct timeval eta_tv;
  41
  42 enum {
  43         Client_created          = 0,
  44         Client_connected        = 1,
  45         Client_started          = 2,
  46         Client_running          = 3,
  47         Client_stopped          = 4,
  48         Client_exited           = 5,
  49 };
  50
  51 static FLIST_HEAD(client_list);
  52 static FLIST_HEAD(eta_list);
  53
  54 static FLIST_HEAD(arg_list);
  55
  56 static struct thread_stat client_ts;
  57 static struct group_run_stats client_gs;
  58 static int sum_stat_clients;
  59 static int sum_stat_nr;
  60
  61 #define FIO_CLIENT_HASH_BITS    7
  62 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  63 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  64 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  65
  66 static void fio_client_add_hash(struct fio_client *client)
  67 {
  68         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  69
  70         bucket &= FIO_CLIENT_HASH_MASK;
  71         flist_add(&client->hash_list, &client_hash[bucket]);
  72 }
  73
  74 static void fio_client_remove_hash(struct fio_client *client)
  75 {
  76         if (!flist_empty(&client->hash_list))
  77                 flist_del_init(&client->hash_list);
  78 }
  79
  80 static void fio_init fio_client_hash_init(void)
  81 {
  82         int i;
  83
  84         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
  85                 INIT_FLIST_HEAD(&client_hash[i]);
  86 }
  87
  88 static struct fio_client *find_client_by_fd(int fd)
  89 {
  90         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
  91         struct fio_client *client;
  92         struct flist_head *entry;
  93
  94         flist_for_each(entry, &client_hash[bucket]) {
  95                 client = flist_entry(entry, struct fio_client, hash_list);
  96
  97                 if (client->fd == fd)
  98                         return client;
  99         }
 100
 101         return NULL;
 102 }
 103
 104 static void remove_client(struct fio_client *client)
 105 {
 106         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 107         flist_del(&client->list);
 108
 109         fio_client_remove_hash(client);
 110
 111         if (!flist_empty(&client->eta_list)) {
 112                 flist_del_init(&client->eta_list);
 113                 fio_client_dec_jobs_eta(client->eta_in_flight, display_thread_status);
 114         }
 115
 116         free(client->hostname);
 117         if (client->argv)
 118                 free(client->argv);
 119         if (client->name)
 120                 free(client->name);
 121
 122         free(client);
 123         nr_clients--;
 124         sum_stat_clients--;
 125 }
 126
 127 static void __fio_client_add_cmd_option(struct fio_client *client,
 128                                         const char *opt)
 129 {
 130         int index;
 131
 132         index = client->argc++;
 133         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 134         client->argv[index] = strdup(opt);
 135         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 136 }
 137
 138 void fio_client_add_cmd_option(void *cookie, const char *opt)
 139 {
 140         struct fio_client *client = cookie;
 141         struct flist_head *entry;
 142
 143         if (!client || !opt)
 144                 return;
 145
 146         __fio_client_add_cmd_option(client, opt);
 147
 148         /*
 149          * Duplicate arguments to shared client group
 150          */
 151         flist_for_each(entry, &arg_list) {
 152                 client = flist_entry(entry, struct fio_client, arg_list);
 153
 154                 __fio_client_add_cmd_option(client, opt);
 155         }
 156 }
 157
 158 struct fio_client *fio_client_add_explicit(const char *hostname, int type,
 159                                            int port)
 160 {
 161         struct fio_client *client;
 162
 163         client = malloc(sizeof(*client));
 164         memset(client, 0, sizeof(*client));
 165
 166         INIT_FLIST_HEAD(&client->list);
 167         INIT_FLIST_HEAD(&client->hash_list);
 168         INIT_FLIST_HEAD(&client->arg_list);
 169         INIT_FLIST_HEAD(&client->eta_list);
 170         INIT_FLIST_HEAD(&client->cmd_list);
 171
 172         client->hostname = strdup(hostname);
 173
 174         if (type == Fio_client_socket)
 175                 client->is_sock = 1;
 176         else {
 177                 int ipv6;
 178
 179                 ipv6 = type == Fio_client_ipv6;
 180                 if (fio_server_parse_host(hostname, &ipv6,
 181                                                 &client->addr.sin_addr,
 182                                                 &client->addr6.sin6_addr))
 183                         goto err;
 184
 185                 client->port = port;
 186         }
 187
 188         client->fd = -1;
 189
 190         __fio_client_add_cmd_option(client, "fio");
 191
 192         flist_add(&client->list, &client_list);
 193         nr_clients++;
 194         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 195         return client;
 196 err:
 197         free(client);
 198         return NULL;
 199 }
 200
 201 int fio_client_add(const char *hostname, void **cookie)
 202 {
 203         struct fio_client *existing = *cookie;
 204         struct fio_client *client;
 205
 206         if (existing) {
 207                 /*
 208                  * We always add our "exec" name as the option, hence 1
 209                  * means empty.
 210                  */
 211                 if (existing->argc == 1)
 212                         flist_add_tail(&existing->arg_list, &arg_list);
 213                 else {
 214                         while (!flist_empty(&arg_list))
 215                                 flist_del_init(arg_list.next);
 216                 }
 217         }
 218
 219         client = malloc(sizeof(*client));
 220         memset(client, 0, sizeof(*client));
 221
 222         INIT_FLIST_HEAD(&client->list);
 223         INIT_FLIST_HEAD(&client->hash_list);
 224         INIT_FLIST_HEAD(&client->arg_list);
 225         INIT_FLIST_HEAD(&client->eta_list);
 226         INIT_FLIST_HEAD(&client->cmd_list);
 227
 228         if (fio_server_parse_string(hostname, &client->hostname,
 229                                         &client->is_sock, &client->port,
 230                                         &client->addr.sin_addr,
 231                                         &client->addr6.sin6_addr,
 232                                         &client->ipv6))
 233                 return -1;
 234
 235         client->fd = -1;
 236
 237         __fio_client_add_cmd_option(client, "fio");
 238
 239         flist_add(&client->list, &client_list);
 240         nr_clients++;
 241         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 242         *cookie = client;
 243         return 0;
 244 }
 245
 246 static int fio_client_connect_ip(struct fio_client *client)
 247 {
 248         struct sockaddr *addr;
 249         fio_socklen_t socklen;
 250         int fd, domain;
 251
 252         if (client->ipv6) {
 253                 client->addr6.sin6_family = AF_INET6;
 254                 client->addr6.sin6_port = htons(client->port);
 255                 domain = AF_INET6;
 256                 addr = (struct sockaddr *) &client->addr6;
 257                 socklen = sizeof(client->addr6);
 258         } else {
 259                 client->addr.sin_family = AF_INET;
 260                 client->addr.sin_port = htons(client->port);
 261                 domain = AF_INET;
 262                 addr = (struct sockaddr *) &client->addr;
 263                 socklen = sizeof(client->addr);
 264         }
 265
 266         fd = socket(domain, SOCK_STREAM, 0);
 267         if (fd < 0) {
 268                 log_err("fio: socket: %s\n", strerror(errno));
 269                 return -1;
 270         }
 271
 272         if (connect(fd, addr, socklen) < 0) {
 273                 log_err("fio: connect: %s\n", strerror(errno));
 274                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 275                                                                 client->port);
 276                 close(fd);
 277                 return -1;
 278         }
 279
 280         return fd;
 281 }
 282
 283 static int fio_client_connect_sock(struct fio_client *client)
 284 {
 285         struct sockaddr_un *addr = &client->addr_un;
 286         fio_socklen_t len;
 287         int fd;
 288
 289         memset(addr, 0, sizeof(*addr));
 290         addr->sun_family = AF_UNIX;
 291         strcpy(addr->sun_path, client->hostname);
 292
 293         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 294         if (fd < 0) {
 295                 log_err("fio: socket: %s\n", strerror(errno));
 296                 return -1;
 297         }
 298
 299         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 300         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 301                 log_err("fio: connect; %s\n", strerror(errno));
 302                 close(fd);
 303                 return -1;
 304         }
 305
 306         return fd;
 307 }
 308
 309 static int fio_client_connect(struct fio_client *client)
 310 {
 311         int fd;
 312
 313         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 314
 315         if (client->is_sock)
 316                 fd = fio_client_connect_sock(client);
 317         else
 318                 fd = fio_client_connect_ip(client);
 319
 320         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 321
 322         if (fd < 0)
 323                 return 1;
 324
 325         client->fd = fd;
 326         fio_client_add_hash(client);
 327         client->state = Client_connected;
 328         return 0;
 329 }
 330
 331 void fio_clients_terminate(void)
 332 {
 333         struct flist_head *entry;
 334         struct fio_client *client;
 335
 336         dprint(FD_NET, "client: terminate clients\n");
 337
 338         flist_for_each(entry, &client_list) {
 339                 client = flist_entry(entry, struct fio_client, list);
 340
 341                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0, NULL);
 342         }
 343 }
 344
 345 static void sig_int(int sig)
 346 {
 347         dprint(FD_NET, "client: got signal %d\n", sig);
 348         fio_clients_terminate();
 349 }
 350
 351 static void client_signal_handler(void)
 352 {
 353         struct sigaction act;
 354
 355         memset(&act, 0, sizeof(act));
 356         act.sa_handler = sig_int;
 357         act.sa_flags = SA_RESTART;
 358         sigaction(SIGINT, &act, NULL);
 359
 360         memset(&act, 0, sizeof(act));
 361         act.sa_handler = sig_int;
 362         act.sa_flags = SA_RESTART;
 363         sigaction(SIGTERM, &act, NULL);
 364 }
 365
 366 static void probe_client(struct fio_client *client)
 367 {
 368         dprint(FD_NET, "client: send probe\n");
 369
 370         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
 371 }
 372
 373 static int send_client_cmd_line(struct fio_client *client)
 374 {
 375         struct cmd_single_line_pdu *cslp;
 376         struct cmd_line_pdu *clp;
 377         unsigned long offset;
 378         unsigned int *lens;
 379         void *pdu;
 380         size_t mem;
 381         int i, ret;
 382
 383         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 384
 385         lens = malloc(client->argc * sizeof(unsigned int));
 386
 387         /*
 388          * Find out how much mem we need
 389          */
 390         for (i = 0, mem = 0; i < client->argc; i++) {
 391                 lens[i] = strlen(client->argv[i]) + 1;
 392                 mem += lens[i];
 393         }
 394
 395         /*
 396          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 397          */
 398         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 399
 400         pdu = malloc(mem);
 401         clp = pdu;
 402         offset = sizeof(*clp);
 403
 404         for (i = 0; i < client->argc; i++) {
 405                 uint16_t arg_len = lens[i];
 406
 407                 cslp = pdu + offset;
 408                 strcpy((char *) cslp->text, client->argv[i]);
 409                 cslp->len = cpu_to_le16(arg_len);
 410                 offset += sizeof(*cslp) + arg_len;
 411         }
 412
 413         free(lens);
 414         clp->lines = cpu_to_le16(client->argc);
 415         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
 416         free(pdu);
 417         return ret;
 418 }
 419
 420 int fio_clients_connect(void)
 421 {
 422         struct fio_client *client;
 423         struct flist_head *entry, *tmp;
 424         int ret;
 425
 426 #ifdef WIN32
 427         WSADATA wsd;
 428         WSAStartup(MAKEWORD(2,2), &wsd);
 429 #endif
 430
 431         dprint(FD_NET, "client: connect all\n");
 432
 433         client_signal_handler();
 434
 435         flist_for_each_safe(entry, tmp, &client_list) {
 436                 client = flist_entry(entry, struct fio_client, list);
 437
 438                 ret = fio_client_connect(client);
 439                 if (ret) {
 440                         remove_client(client);
 441                         continue;
 442                 }
 443
 444                 probe_client(client);
 445
 446                 if (client->argc > 1)
 447                         send_client_cmd_line(client);
 448         }
 449
 450         return !nr_clients;
 451 }
 452
 453 /*
 454  * Send file contents to server backend. We could use sendfile(), but to remain
 455  * more portable lets just read/write the darn thing.
 456  */
 457 static int fio_client_send_ini(struct fio_client *client, const char *filename)
 458 {
 459         struct stat sb;
 460         char *p, *buf;
 461         off_t len;
 462         int fd, ret;
 463
 464         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 465
 466         fd = open(filename, O_RDONLY);
 467         if (fd < 0) {
 468                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 469                 return 1;
 470         }
 471
 472         if (fstat(fd, &sb) < 0) {
 473                 log_err("fio: job file stat: %s\n", strerror(errno));
 474                 close(fd);
 475                 return 1;
 476         }
 477
 478         buf = malloc(sb.st_size);
 479
 480         len = sb.st_size;
 481         p = buf;
 482         do {
 483                 ret = read(fd, p, len);
 484                 if (ret > 0) {
 485                         len -= ret;
 486                         if (!len)
 487                                 break;
 488                         p += ret;
 489                         continue;
 490                 } else if (!ret)
 491                         break;
 492                 else if (errno == EAGAIN || errno == EINTR)
 493                         continue;
 494         } while (1);
 495
 496         if (len) {
 497                 log_err("fio: failed reading job file %s\n", filename);
 498                 close(fd);
 499                 free(buf);
 500                 return 1;
 501         }
 502
 503         client->sent_job = 1;
 504         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size, 0);
 505         free(buf);
 506         close(fd);
 507         return ret;
 508 }
 509
 510 int fio_clients_send_ini(const char *filename)
 511 {
 512         struct fio_client *client;
 513         struct flist_head *entry, *tmp;
 514
 515         flist_for_each_safe(entry, tmp, &client_list) {
 516                 client = flist_entry(entry, struct fio_client, list);
 517
 518                 if (fio_client_send_ini(client, filename))
 519                         remove_client(client);
 520
 521                 client->sent_job = 1;
 522         }
 523
 524         return !nr_clients;
 525 }
 526
 527 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 528 {
 529         dst->max_val    = le64_to_cpu(src->max_val);
 530         dst->min_val    = le64_to_cpu(src->min_val);
 531         dst->samples    = le64_to_cpu(src->samples);
 532
 533         /*
 534          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 535          */
 536         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 537         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 538 }
 539
 540 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 541 {
 542         int i, j;
 543
 544         dst->error      = le32_to_cpu(src->error);
 545         dst->groupid    = le32_to_cpu(src->groupid);
 546         dst->pid        = le32_to_cpu(src->pid);
 547         dst->members    = le32_to_cpu(src->members);
 548
 549         for (i = 0; i < 2; i++) {
 550                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 551                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 552                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 553                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 554         }
 555
 556         dst->usr_time           = le64_to_cpu(src->usr_time);
 557         dst->sys_time           = le64_to_cpu(src->sys_time);
 558         dst->ctx                = le64_to_cpu(src->ctx);
 559         dst->minf               = le64_to_cpu(src->minf);
 560         dst->majf               = le64_to_cpu(src->majf);
 561         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 562
 563         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 564                 fio_fp64_t *fps = &src->percentile_list[i];
 565                 fio_fp64_t *fpd = &dst->percentile_list[i];
 566
 567                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 568         }
 569
 570         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 571                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 572                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 573                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 574         }
 575
 576         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 577                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 578                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 579         }
 580
 581         for (i = 0; i < 2; i++)
 582                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 583                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 584
 585         for (i = 0; i < 3; i++) {
 586                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 587                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 588         }
 589
 590         dst->total_submit       = le64_to_cpu(src->total_submit);
 591         dst->total_complete     = le64_to_cpu(src->total_complete);
 592
 593         for (i = 0; i < 2; i++) {
 594                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 595                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 596         }
 597
 598         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 599         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 600         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 601         dst->first_error        = le32_to_cpu(src->first_error);
 602         dst->kb_base            = le32_to_cpu(src->kb_base);
 603 }
 604
 605 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 606 {
 607         int i;
 608
 609         for (i = 0; i < 2; i++) {
 610                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 611                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 612                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 613                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 614                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 615                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 616         }
 617
 618         dst->kb_base    = le32_to_cpu(src->kb_base);
 619         dst->groupid    = le32_to_cpu(src->groupid);
 620 }
 621
 622 static void handle_ts(struct fio_net_cmd *cmd)
 623 {
 624         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 625
 626         convert_ts(&p->ts, &p->ts);
 627         convert_gs(&p->rs, &p->rs);
 628
 629         show_thread_status(&p->ts, &p->rs);
 630
 631         if (sum_stat_clients == 1)
 632                 return;
 633
 634         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 635         sum_group_stats(&client_gs, &p->rs);
 636
 637         client_ts.members++;
 638         client_ts.groupid = p->ts.groupid;
 639
 640         if (++sum_stat_nr == sum_stat_clients) {
 641                 strcpy(client_ts.name, "All clients");
 642                 show_thread_status(&client_ts, &client_gs);
 643         }
 644 }
 645
 646 static void handle_gs(struct fio_net_cmd *cmd)
 647 {
 648         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 649
 650         convert_gs(gs, gs);
 651         show_group_stats(gs);
 652 }
 653
 654 static void convert_agg(struct disk_util_agg *agg)
 655 {
 656         int i;
 657
 658         for (i = 0; i < 2; i++) {
 659                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 660                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 661                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 662                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 663         }
 664
 665         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 666         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 667         agg->slavecount         = le32_to_cpu(agg->slavecount);
 668         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 669 }
 670
 671 static void convert_dus(struct disk_util_stat *dus)
 672 {
 673         int i;
 674
 675         for (i = 0; i < 2; i++) {
 676                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
 677                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
 678                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
 679                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
 680         }
 681
 682         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
 683         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
 684         dus->msec               = le64_to_cpu(dus->msec);
 685 }
 686
 687 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
 688 {
 689         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
 690         const char *buf = (const char *) pdu->buf;
 691         const char *name;
 692         int fio_unused ret;
 693
 694         name = client->name ? client->name : client->hostname;
 695
 696         if (!client->skip_newline)
 697                 fprintf(f_out, "<%s> ", name);
 698         ret = fwrite(buf, pdu->buf_len, 1, f_out);
 699         fflush(f_out);
 700         client->skip_newline = strchr(buf, '\n') == NULL;
 701 }
 702
 703
 704 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 705 {
 706         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 707
 708         convert_dus(&du->dus);
 709         convert_agg(&du->agg);
 710
 711         if (!client->disk_stats_shown) {
 712                 client->disk_stats_shown = 1;
 713                 log_info("\nDisk stats (read/write):\n");
 714         }
 715
 716         print_disk_util(&du->dus, &du->agg, terse_output);
 717 }
 718
 719 void fio_client_convert_jobs_eta(struct jobs_eta *je)
 720 {
 721         int i;
 722
 723         je->nr_running          = le32_to_cpu(je->nr_running);
 724         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 725         je->nr_pending          = le32_to_cpu(je->nr_pending);
 726         je->files_open          = le32_to_cpu(je->files_open);
 727
 728         for (i = 0; i < 2; i++) {
 729                 je->m_rate[i]           = le32_to_cpu(je->m_rate[i]);
 730                 je->t_rate[i]           = le32_to_cpu(je->t_rate[i]);
 731                 je->m_iops[i]           = le32_to_cpu(je->m_iops[i]);
 732                 je->t_iops[i]           = le32_to_cpu(je->t_iops[i]);
 733                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 734                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 735         }
 736
 737         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 738         je->eta_sec             = le64_to_cpu(je->eta_sec);
 739 }
 740
 741 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 742 {
 743         int i;
 744
 745         dst->nr_running         += je->nr_running;
 746         dst->nr_ramp            += je->nr_ramp;
 747         dst->nr_pending         += je->nr_pending;
 748         dst->files_open         += je->files_open;
 749
 750         for (i = 0; i < 2; i++) {
 751                 dst->m_rate[i]  += je->m_rate[i];
 752                 dst->t_rate[i]  += je->t_rate[i];
 753                 dst->m_iops[i]  += je->m_iops[i];
 754                 dst->t_iops[i]  += je->t_iops[i];
 755                 dst->rate[i]    += je->rate[i];
 756                 dst->iops[i]    += je->iops[i];
 757         }
 758
 759         dst->elapsed_sec        += je->elapsed_sec;
 760
 761         if (je->eta_sec > dst->eta_sec)
 762                 dst->eta_sec = je->eta_sec;
 763 }
 764
 765 void fio_client_dec_jobs_eta(struct client_eta *eta, void (*fn)(struct jobs_eta *))
 766 {
 767         if (!--eta->pending) {
 768                 fn(&eta->eta);
 769                 free(eta);
 770         }
 771 }
 772
 773 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 774 {
 775         struct fio_net_int_cmd *icmd = NULL;
 776         struct flist_head *entry;
 777
 778         flist_for_each(entry, &client->cmd_list) {
 779                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
 780
 781                 if (cmd->tag == (uintptr_t) icmd)
 782                         break;
 783
 784                 icmd = NULL;
 785         }
 786
 787         if (!icmd) {
 788                 log_err("fio: client: unable to find matching tag\n");
 789                 return;
 790         }
 791
 792         flist_del(&icmd->list);
 793         cmd->tag = icmd->saved_tag;
 794         free(icmd);
 795 }
 796
 797 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 798 {
 799         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 800         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 801
 802         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 803
 804         assert(client->eta_in_flight == eta);
 805
 806         client->eta_in_flight = NULL;
 807         flist_del_init(&client->eta_list);
 808
 809         fio_client_convert_jobs_eta(je);
 810         fio_client_sum_jobs_eta(&eta->eta, je);
 811         fio_client_dec_jobs_eta(eta, display_thread_status);
 812 }
 813
 814 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 815 {
 816         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 817         const char *os, *arch;
 818         char bit[16];
 819
 820         os = fio_get_os_string(probe->os);
 821         if (!os)
 822                 os = "unknown";
 823
 824         arch = fio_get_arch_string(probe->arch);
 825         if (!arch)
 826                 os = "unknown";
 827
 828         sprintf(bit, "%d-bit", probe->bpp * 8);
 829
 830         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
 831                 probe->hostname, probe->bigendian, bit, os, arch,
 832                 probe->fio_major, probe->fio_minor, probe->fio_patch);
 833
 834         if (!client->name)
 835                 client->name = strdup((char *) probe->hostname);
 836 }
 837
 838 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
 839 {
 840         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 841
 842         client->state = Client_started;
 843         client->jobs = le32_to_cpu(pdu->jobs);
 844 }
 845
 846 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
 847 {
 848         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
 849
 850         client->state = Client_stopped;
 851         client->error = le32_to_cpu(pdu->error);
 852
 853         if (client->error)
 854                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
 855 }
 856
 857 static void convert_text(struct fio_net_cmd *cmd)
 858 {
 859         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
 860
 861         pdu->level      = le32_to_cpu(pdu->level);
 862         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
 863         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
 864         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
 865 }
 866
 867 int fio_handle_client(struct fio_client *client, struct client_ops *ops)
 868 {
 869         struct fio_net_cmd *cmd;
 870
 871         dprint(FD_NET, "client: handle %s\n", client->hostname);
 872
 873         cmd = fio_net_recv_cmd(client->fd);
 874         if (!cmd)
 875                 return 0;
 876
 877         dprint(FD_NET, "client: got cmd op %s from %s\n",
 878                                 fio_server_op(cmd->opcode), client->hostname);
 879
 880         switch (cmd->opcode) {
 881         case FIO_NET_CMD_QUIT:
 882                 if (ops->quit)
 883                         ops->quit(client);
 884                 remove_client(client);
 885                 free(cmd);
 886                 break;
 887         case FIO_NET_CMD_TEXT:
 888                 convert_text(cmd);
 889                 ops->text_op(client, cmd);
 890                 free(cmd);
 891                 break;
 892         case FIO_NET_CMD_DU:
 893                 ops->disk_util(client, cmd);
 894                 free(cmd);
 895                 break;
 896         case FIO_NET_CMD_TS:
 897                 ops->thread_status(cmd);
 898                 free(cmd);
 899                 break;
 900         case FIO_NET_CMD_GS:
 901                 ops->group_stats(cmd);
 902                 free(cmd);
 903                 break;
 904         case FIO_NET_CMD_ETA:
 905                 remove_reply_cmd(client, cmd);
 906                 ops->eta(client, cmd);
 907                 free(cmd);
 908                 break;
 909         case FIO_NET_CMD_PROBE:
 910                 remove_reply_cmd(client, cmd);
 911                 ops->probe(client, cmd);
 912                 free(cmd);
 913                 break;
 914         case FIO_NET_CMD_RUN:
 915                 client->state = Client_running;
 916                 free(cmd);
 917                 break;
 918         case FIO_NET_CMD_START:
 919                 handle_start(client, cmd);
 920                 free(cmd);
 921                 break;
 922         case FIO_NET_CMD_STOP:
 923                 handle_stop(client, cmd);
 924                 free(cmd);
 925                 break;
 926         case FIO_NET_CMD_ADD_JOB:
 927                 if (ops->add_job)
 928                         ops->add_job(client, cmd);
 929                 free(cmd);
 930                 break;
 931         default:
 932                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
 933                 free(cmd);
 934                 break;
 935         }
 936
 937         return 1;
 938 }
 939
 940 static void request_client_etas(void)
 941 {
 942         struct fio_client *client;
 943         struct flist_head *entry;
 944         struct client_eta *eta;
 945         int skipped = 0;
 946
 947         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
 948
 949         eta = malloc(sizeof(*eta));
 950         memset(&eta->eta, 0, sizeof(eta->eta));
 951         eta->pending = nr_clients;
 952
 953         flist_for_each(entry, &client_list) {
 954                 client = flist_entry(entry, struct fio_client, list);
 955
 956                 if (!flist_empty(&client->eta_list)) {
 957                         skipped++;
 958                         continue;
 959                 }
 960                 if (client->state != Client_running)
 961                         continue;
 962
 963                 assert(!client->eta_in_flight);
 964                 flist_add_tail(&client->eta_list, &eta_list);
 965                 client->eta_in_flight = eta;
 966                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
 967                                         (uintptr_t) eta, &client->cmd_list);
 968         }
 969
 970         while (skipped--)
 971                 fio_client_dec_jobs_eta(eta, display_thread_status);
 972
 973         dprint(FD_NET, "client: requested eta tag %p\n", eta);
 974 }
 975
 976 static int client_check_cmd_timeout(struct fio_client *client,
 977                                     struct timeval *now)
 978 {
 979         struct fio_net_int_cmd *cmd;
 980         struct flist_head *entry, *tmp;
 981         int ret = 0;
 982
 983         flist_for_each_safe(entry, tmp, &client->cmd_list) {
 984                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
 985
 986                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
 987                         continue;
 988
 989                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
 990                                                 fio_server_op(cmd->cmd.opcode));
 991                 flist_del(&cmd->list);
 992                 free(cmd);
 993                 ret = 1;
 994         }
 995
 996         return flist_empty(&client->cmd_list) && ret;
 997 }
 998
 999 static int fio_client_timed_out(struct client_ops *ops)
1000 {
1001         struct fio_client *client;
1002         struct flist_head *entry, *tmp;
1003         struct timeval tv;
1004         int ret = 0;
1005
1006         gettimeofday(&tv, NULL);
1007
1008         flist_for_each_safe(entry, tmp, &client_list) {
1009                 client = flist_entry(entry, struct fio_client, list);
1010
1011                 if (flist_empty(&client->cmd_list))
1012                         continue;
1013
1014                 if (!client_check_cmd_timeout(client, &tv))
1015                         continue;
1016
1017                 if (ops->timed_out)
1018                         ops->timed_out(client);
1019                 else
1020                         log_err("fio: client %s timed out\n", client->hostname);
1021
1022                 remove_client(client);
1023                 ret = 1;
1024         }
1025
1026         return ret;
1027 }
1028
1029 int fio_handle_clients(struct client_ops *ops)
1030 {
1031         struct pollfd *pfds;
1032         int i, ret = 0, retval = 0;
1033
1034         gettimeofday(&eta_tv, NULL);
1035
1036         pfds = malloc(nr_clients * sizeof(struct pollfd));
1037
1038         sum_stat_clients = nr_clients;
1039         init_thread_stat(&client_ts);
1040         init_group_run_stat(&client_gs);
1041
1042         while (!exit_backend && nr_clients) {
1043                 struct flist_head *entry, *tmp;
1044                 struct fio_client *client;
1045
1046                 i = 0;
1047                 flist_for_each_safe(entry, tmp, &client_list) {
1048                         client = flist_entry(entry, struct fio_client, list);
1049
1050                         if (!client->sent_job && !ops->stay_connected &&
1051                             flist_empty(&client->cmd_list)) {
1052                                 remove_client(client);
1053                                 continue;
1054                         }
1055
1056                         pfds[i].fd = client->fd;
1057                         pfds[i].events = POLLIN;
1058                         i++;
1059                 }
1060
1061                 if (!nr_clients)
1062                         break;
1063
1064                 assert(i == nr_clients);
1065
1066                 do {
1067                         struct timeval tv;
1068
1069                         gettimeofday(&tv, NULL);
1070                         if (mtime_since(&eta_tv, &tv) >= 900) {
1071                                 request_client_etas();
1072                                 memcpy(&eta_tv, &tv, sizeof(tv));
1073
1074                                 if (fio_client_timed_out(ops))
1075                                         break;
1076                         }
1077
1078                         ret = poll(pfds, nr_clients, 100);
1079                         if (ret < 0) {
1080                                 if (errno == EINTR)
1081                                         continue;
1082                                 log_err("fio: poll clients: %s\n", strerror(errno));
1083                                 break;
1084                         } else if (!ret)
1085                                 continue;
1086                 } while (ret <= 0);
1087
1088                 for (i = 0; i < nr_clients; i++) {
1089                         if (!(pfds[i].revents & POLLIN))
1090                                 continue;
1091
1092                         client = find_client_by_fd(pfds[i].fd);
1093                         if (!client) {
1094                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1095                                 continue;
1096                         }
1097                         if (!fio_handle_client(client, ops)) {
1098                                 log_info("client: host=%s disconnected\n",
1099                                                 client->hostname);
1100                                 remove_client(client);
1101                                 retval = 1;
1102                         } else if (client->error)
1103                                 retval = 1;
1104                 }
1105         }
1106
1107         free(pfds);
1108         return retval;
1109 }