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 "server.h"
  20 #include "flist.h"
  21 #include "hash.h"
  22
  23 struct client_eta {
  24         struct jobs_eta eta;
  25         unsigned int pending;
  26 };
  27
  28 struct fio_client {
  29         struct flist_head list;
  30         struct flist_head hash_list;
  31         struct flist_head arg_list;
  32         union {
  33                 struct sockaddr_in addr;
  34                 struct sockaddr_in6 addr6;
  35                 struct sockaddr_un addr_un;
  36         };
  37         char *hostname;
  38         int port;
  39         int fd;
  40         unsigned int refs;
  41
  42         char *name;
  43
  44         int state;
  45
  46         int skip_newline;
  47         int is_sock;
  48         int disk_stats_shown;
  49         unsigned int jobs;
  50         int error;
  51         int ipv6;
  52         int sent_job;
  53
  54         struct flist_head eta_list;
  55         struct client_eta *eta_in_flight;
  56
  57         struct flist_head cmd_list;
  58
  59         uint16_t argc;
  60         char **argv;
  61 };
  62
  63 static struct timeval eta_tv;
  64
  65 enum {
  66         Client_created          = 0,
  67         Client_connected        = 1,
  68         Client_started          = 2,
  69         Client_running          = 3,
  70         Client_stopped          = 4,
  71         Client_exited           = 5,
  72 };
  73
  74 static FLIST_HEAD(client_list);
  75 static FLIST_HEAD(eta_list);
  76
  77 static FLIST_HEAD(arg_list);
  78
  79 static struct thread_stat client_ts;
  80 static struct group_run_stats client_gs;
  81 static int sum_stat_clients;
  82 static int sum_stat_nr;
  83
  84 #define FIO_CLIENT_HASH_BITS    7
  85 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  86 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  87 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  88
  89 static int handle_client(struct fio_client *client);
  90 static void dec_jobs_eta(struct client_eta *eta);
  91
  92 static void fio_client_add_hash(struct fio_client *client)
  93 {
  94         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  95
  96         bucket &= FIO_CLIENT_HASH_MASK;
  97         flist_add(&client->hash_list, &client_hash[bucket]);
  98 }
  99
 100 static void fio_client_remove_hash(struct fio_client *client)
 101 {
 102         if (!flist_empty(&client->hash_list))
 103                 flist_del_init(&client->hash_list);
 104 }
 105
 106 static void fio_init fio_client_hash_init(void)
 107 {
 108         int i;
 109
 110         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
 111                 INIT_FLIST_HEAD(&client_hash[i]);
 112 }
 113
 114 static struct fio_client *find_client_by_fd(int fd)
 115 {
 116         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
 117         struct fio_client *client;
 118         struct flist_head *entry;
 119
 120         flist_for_each(entry, &client_hash[bucket]) {
 121                 client = flist_entry(entry, struct fio_client, hash_list);
 122
 123                 if (client->fd == fd) {
 124                         client->refs++;
 125                         return client;
 126                 }
 127         }
 128
 129         return NULL;
 130 }
 131
 132 static void remove_client(struct fio_client *client)
 133 {
 134         assert(client->refs);
 135
 136         if (--client->refs)
 137                 return;
 138
 139         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 140         flist_del(&client->list);
 141
 142         fio_client_remove_hash(client);
 143
 144         if (!flist_empty(&client->eta_list)) {
 145                 flist_del_init(&client->eta_list);
 146                 dec_jobs_eta(client->eta_in_flight);
 147         }
 148
 149         free(client->hostname);
 150         if (client->argv)
 151                 free(client->argv);
 152         if (client->name)
 153                 free(client->name);
 154
 155         free(client);
 156         nr_clients--;
 157         sum_stat_clients--;
 158 }
 159
 160 static void put_client(struct fio_client *client)
 161 {
 162         remove_client(client);
 163 }
 164
 165 static void __fio_client_add_cmd_option(struct fio_client *client,
 166                                         const char *opt)
 167 {
 168         int index;
 169
 170         index = client->argc++;
 171         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 172         client->argv[index] = strdup(opt);
 173         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 174 }
 175
 176 void fio_client_add_cmd_option(void *cookie, const char *opt)
 177 {
 178         struct fio_client *client = cookie;
 179         struct flist_head *entry;
 180
 181         if (!client || !opt)
 182                 return;
 183
 184         __fio_client_add_cmd_option(client, opt);
 185
 186         /*
 187          * Duplicate arguments to shared client group
 188          */
 189         flist_for_each(entry, &arg_list) {
 190                 client = flist_entry(entry, struct fio_client, arg_list);
 191
 192                 __fio_client_add_cmd_option(client, opt);
 193         }
 194 }
 195
 196 int fio_client_add(const char *hostname, void **cookie)
 197 {
 198         struct fio_client *existing = *cookie;
 199         struct fio_client *client;
 200
 201         if (existing) {
 202                 /*
 203                  * We always add our "exec" name as the option, hence 1
 204                  * means empty.
 205                  */
 206                 if (existing->argc == 1)
 207                         flist_add_tail(&existing->arg_list, &arg_list);
 208                 else {
 209                         while (!flist_empty(&arg_list))
 210                                 flist_del_init(arg_list.next);
 211                 }
 212         }
 213
 214         client = malloc(sizeof(*client));
 215         memset(client, 0, sizeof(*client));
 216
 217         INIT_FLIST_HEAD(&client->list);
 218         INIT_FLIST_HEAD(&client->hash_list);
 219         INIT_FLIST_HEAD(&client->arg_list);
 220         INIT_FLIST_HEAD(&client->eta_list);
 221         INIT_FLIST_HEAD(&client->cmd_list);
 222
 223         if (fio_server_parse_string(hostname, &client->hostname,
 224                                         &client->is_sock, &client->port,
 225                                         &client->addr.sin_addr,
 226                                         &client->addr6.sin6_addr,
 227                                         &client->ipv6))
 228                 return -1;
 229
 230         client->fd = -1;
 231         client->refs = 1;
 232
 233         __fio_client_add_cmd_option(client, "fio");
 234
 235         flist_add(&client->list, &client_list);
 236         nr_clients++;
 237         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 238         *cookie = client;
 239         return 0;
 240 }
 241
 242 static int fio_client_connect_ip(struct fio_client *client)
 243 {
 244         struct sockaddr *addr;
 245         fio_socklen_t socklen;
 246         int fd, domain;
 247
 248         if (client->ipv6) {
 249                 client->addr6.sin6_family = AF_INET6;
 250                 client->addr6.sin6_port = htons(client->port);
 251                 domain = AF_INET6;
 252                 addr = (struct sockaddr *) &client->addr6;
 253                 socklen = sizeof(client->addr6);
 254         } else {
 255                 client->addr.sin_family = AF_INET;
 256                 client->addr.sin_port = htons(client->port);
 257                 domain = AF_INET;
 258                 addr = (struct sockaddr *) &client->addr;
 259                 socklen = sizeof(client->addr);
 260         }
 261
 262         fd = socket(domain, SOCK_STREAM, 0);
 263         if (fd < 0) {
 264                 log_err("fio: socket: %s\n", strerror(errno));
 265                 return -1;
 266         }
 267
 268         if (connect(fd, addr, socklen) < 0) {
 269                 log_err("fio: connect: %s\n", strerror(errno));
 270                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 271                                                                 client->port);
 272                 close(fd);
 273                 return -1;
 274         }
 275
 276         return fd;
 277 }
 278
 279 static int fio_client_connect_sock(struct fio_client *client)
 280 {
 281         struct sockaddr_un *addr = &client->addr_un;
 282         fio_socklen_t len;
 283         int fd;
 284
 285         memset(addr, 0, sizeof(*addr));
 286         addr->sun_family = AF_UNIX;
 287         strcpy(addr->sun_path, client->hostname);
 288
 289         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 290         if (fd < 0) {
 291                 log_err("fio: socket: %s\n", strerror(errno));
 292                 return -1;
 293         }
 294
 295         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 296         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 297                 log_err("fio: connect; %s\n", strerror(errno));
 298                 close(fd);
 299                 return -1;
 300         }
 301
 302         return fd;
 303 }
 304
 305 static int fio_client_connect(struct fio_client *client)
 306 {
 307         int fd;
 308
 309         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 310
 311         if (client->is_sock)
 312                 fd = fio_client_connect_sock(client);
 313         else
 314                 fd = fio_client_connect_ip(client);
 315
 316         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 317
 318         if (fd < 0)
 319                 return 1;
 320
 321         client->fd = fd;
 322         fio_client_add_hash(client);
 323         client->state = Client_connected;
 324         return 0;
 325 }
 326
 327 void fio_clients_terminate(void)
 328 {
 329         struct flist_head *entry;
 330         struct fio_client *client;
 331
 332         dprint(FD_NET, "client: terminate clients\n");
 333
 334         flist_for_each(entry, &client_list) {
 335                 client = flist_entry(entry, struct fio_client, list);
 336
 337                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0, NULL);
 338         }
 339 }
 340
 341 static void sig_int(int sig)
 342 {
 343         dprint(FD_NET, "client: got signal %d\n", sig);
 344         fio_clients_terminate();
 345 }
 346
 347 static void client_signal_handler(void)
 348 {
 349         struct sigaction act;
 350
 351         memset(&act, 0, sizeof(act));
 352         act.sa_handler = sig_int;
 353         act.sa_flags = SA_RESTART;
 354         sigaction(SIGINT, &act, NULL);
 355
 356         memset(&act, 0, sizeof(act));
 357         act.sa_handler = sig_int;
 358         act.sa_flags = SA_RESTART;
 359         sigaction(SIGTERM, &act, NULL);
 360 }
 361
 362 static void probe_client(struct fio_client *client)
 363 {
 364         dprint(FD_NET, "client: send probe\n");
 365
 366         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
 367 }
 368
 369 static int send_client_cmd_line(struct fio_client *client)
 370 {
 371         struct cmd_single_line_pdu *cslp;
 372         struct cmd_line_pdu *clp;
 373         unsigned long offset;
 374         unsigned int *lens;
 375         void *pdu;
 376         size_t mem;
 377         int i, ret;
 378
 379         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 380
 381         lens = malloc(client->argc * sizeof(unsigned int));
 382
 383         /*
 384          * Find out how much mem we need
 385          */
 386         for (i = 0, mem = 0; i < client->argc; i++) {
 387                 lens[i] = strlen(client->argv[i]) + 1;
 388                 mem += lens[i];
 389         }
 390
 391         /*
 392          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 393          */
 394         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 395
 396         pdu = malloc(mem);
 397         clp = pdu;
 398         offset = sizeof(*clp);
 399
 400         for (i = 0; i < client->argc; i++) {
 401                 uint16_t arg_len = lens[i];
 402
 403                 cslp = pdu + offset;
 404                 strcpy((char *) cslp->text, client->argv[i]);
 405                 cslp->len = cpu_to_le16(arg_len);
 406                 offset += sizeof(*cslp) + arg_len;
 407         }
 408
 409         free(lens);
 410         clp->lines = cpu_to_le16(client->argc);
 411         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
 412         free(pdu);
 413         return ret;
 414 }
 415
 416 int fio_clients_connect(void)
 417 {
 418         struct fio_client *client;
 419         struct flist_head *entry, *tmp;
 420         int ret;
 421
 422 #ifdef WIN32
 423         WSADATA wsd;
 424         WSAStartup(MAKEWORD(2,2), &wsd);
 425 #endif
 426
 427         dprint(FD_NET, "client: connect all\n");
 428
 429         client_signal_handler();
 430
 431         flist_for_each_safe(entry, tmp, &client_list) {
 432                 client = flist_entry(entry, struct fio_client, list);
 433
 434                 ret = fio_client_connect(client);
 435                 if (ret) {
 436                         remove_client(client);
 437                         continue;
 438                 }
 439
 440                 probe_client(client);
 441
 442                 if (client->argc > 1)
 443                         send_client_cmd_line(client);
 444         }
 445
 446         return !nr_clients;
 447 }
 448
 449 /*
 450  * Send file contents to server backend. We could use sendfile(), but to remain
 451  * more portable lets just read/write the darn thing.
 452  */
 453 static int fio_client_send_ini(struct fio_client *client, const char *filename)
 454 {
 455         struct stat sb;
 456         char *p, *buf;
 457         off_t len;
 458         int fd, ret;
 459
 460         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 461
 462         fd = open(filename, O_RDONLY);
 463         if (fd < 0) {
 464                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 465                 return 1;
 466         }
 467
 468         if (fstat(fd, &sb) < 0) {
 469                 log_err("fio: job file stat: %s\n", strerror(errno));
 470                 close(fd);
 471                 return 1;
 472         }
 473
 474         buf = malloc(sb.st_size);
 475
 476         len = sb.st_size;
 477         p = buf;
 478         do {
 479                 ret = read(fd, p, len);
 480                 if (ret > 0) {
 481                         len -= ret;
 482                         if (!len)
 483                                 break;
 484                         p += ret;
 485                         continue;
 486                 } else if (!ret)
 487                         break;
 488                 else if (errno == EAGAIN || errno == EINTR)
 489                         continue;
 490         } while (1);
 491
 492         if (len) {
 493                 log_err("fio: failed reading job file %s\n", filename);
 494                 close(fd);
 495                 free(buf);
 496                 return 1;
 497         }
 498
 499         client->sent_job = 1;
 500         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size, 0);
 501         free(buf);
 502         close(fd);
 503         return ret;
 504 }
 505
 506 int fio_clients_send_ini(const char *filename)
 507 {
 508         struct fio_client *client;
 509         struct flist_head *entry, *tmp;
 510
 511         flist_for_each_safe(entry, tmp, &client_list) {
 512                 client = flist_entry(entry, struct fio_client, list);
 513
 514                 if (fio_client_send_ini(client, filename))
 515                         remove_client(client);
 516
 517                 client->sent_job = 1;
 518         }
 519
 520         return !nr_clients;
 521 }
 522
 523 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 524 {
 525         dst->max_val    = le64_to_cpu(src->max_val);
 526         dst->min_val    = le64_to_cpu(src->min_val);
 527         dst->samples    = le64_to_cpu(src->samples);
 528
 529         /*
 530          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 531          */
 532         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 533         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 534 }
 535
 536 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 537 {
 538         int i, j;
 539
 540         dst->error      = le32_to_cpu(src->error);
 541         dst->groupid    = le32_to_cpu(src->groupid);
 542         dst->pid        = le32_to_cpu(src->pid);
 543         dst->members    = le32_to_cpu(src->members);
 544
 545         for (i = 0; i < 2; i++) {
 546                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 547                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 548                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 549                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 550         }
 551
 552         dst->usr_time           = le64_to_cpu(src->usr_time);
 553         dst->sys_time           = le64_to_cpu(src->sys_time);
 554         dst->ctx                = le64_to_cpu(src->ctx);
 555         dst->minf               = le64_to_cpu(src->minf);
 556         dst->majf               = le64_to_cpu(src->majf);
 557         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 558
 559         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 560                 fio_fp64_t *fps = &src->percentile_list[i];
 561                 fio_fp64_t *fpd = &dst->percentile_list[i];
 562
 563                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 564         }
 565
 566         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 567                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 568                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 569                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 570         }
 571
 572         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 573                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 574                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 575         }
 576
 577         for (i = 0; i < 2; i++)
 578                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 579                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 580
 581         for (i = 0; i < 3; i++) {
 582                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 583                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 584         }
 585
 586         dst->total_submit       = le64_to_cpu(src->total_submit);
 587         dst->total_complete     = le64_to_cpu(src->total_complete);
 588
 589         for (i = 0; i < 2; i++) {
 590                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 591                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 592         }
 593
 594         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 595         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 596         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 597         dst->first_error        = le32_to_cpu(src->first_error);
 598         dst->kb_base            = le32_to_cpu(src->kb_base);
 599 }
 600
 601 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 602 {
 603         int i;
 604
 605         for (i = 0; i < 2; i++) {
 606                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 607                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 608                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 609                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 610                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 611                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 612         }
 613
 614         dst->kb_base    = le32_to_cpu(src->kb_base);
 615         dst->groupid    = le32_to_cpu(src->groupid);
 616 }
 617
 618 static void handle_ts(struct fio_net_cmd *cmd)
 619 {
 620         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 621
 622         convert_ts(&p->ts, &p->ts);
 623         convert_gs(&p->rs, &p->rs);
 624
 625         show_thread_status(&p->ts, &p->rs);
 626
 627         if (sum_stat_clients == 1)
 628                 return;
 629
 630         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 631         sum_group_stats(&client_gs, &p->rs);
 632
 633         client_ts.members++;
 634         client_ts.groupid = p->ts.groupid;
 635
 636         if (++sum_stat_nr == sum_stat_clients) {
 637                 strcpy(client_ts.name, "All clients");
 638                 show_thread_status(&client_ts, &client_gs);
 639         }
 640 }
 641
 642 static void handle_gs(struct fio_net_cmd *cmd)
 643 {
 644         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 645
 646         convert_gs(gs, gs);
 647         show_group_stats(gs);
 648 }
 649
 650 static void convert_agg(struct disk_util_agg *agg)
 651 {
 652         int i;
 653
 654         for (i = 0; i < 2; i++) {
 655                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 656                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 657                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 658                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 659         }
 660
 661         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 662         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 663         agg->slavecount         = le32_to_cpu(agg->slavecount);
 664         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 665 }
 666
 667 static void convert_dus(struct disk_util_stat *dus)
 668 {
 669         int i;
 670
 671         for (i = 0; i < 2; i++) {
 672                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
 673                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
 674                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
 675                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
 676         }
 677
 678         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
 679         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
 680         dus->msec               = le64_to_cpu(dus->msec);
 681 }
 682
 683 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 684 {
 685         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 686
 687         convert_dus(&du->dus);
 688         convert_agg(&du->agg);
 689
 690         if (!client->disk_stats_shown) {
 691                 client->disk_stats_shown = 1;
 692                 log_info("\nDisk stats (read/write):\n");
 693         }
 694
 695         print_disk_util(&du->dus, &du->agg, terse_output);
 696 }
 697
 698 static void convert_jobs_eta(struct jobs_eta *je)
 699 {
 700         int i;
 701
 702         je->nr_running          = le32_to_cpu(je->nr_running);
 703         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 704         je->nr_pending          = le32_to_cpu(je->nr_pending);
 705         je->files_open          = le32_to_cpu(je->files_open);
 706         je->m_rate              = le32_to_cpu(je->m_rate);
 707         je->t_rate              = le32_to_cpu(je->t_rate);
 708         je->m_iops              = le32_to_cpu(je->m_iops);
 709         je->t_iops              = le32_to_cpu(je->t_iops);
 710
 711         for (i = 0; i < 2; i++) {
 712                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 713                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 714         }
 715
 716         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 717         je->eta_sec             = le64_to_cpu(je->eta_sec);
 718 }
 719
 720 static void sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 721 {
 722         int i;
 723
 724         dst->nr_running         += je->nr_running;
 725         dst->nr_ramp            += je->nr_ramp;
 726         dst->nr_pending         += je->nr_pending;
 727         dst->files_open         += je->files_open;
 728         dst->m_rate             += je->m_rate;
 729         dst->t_rate             += je->t_rate;
 730         dst->m_iops             += je->m_iops;
 731         dst->t_iops             += je->t_iops;
 732
 733         for (i = 0; i < 2; i++) {
 734                 dst->rate[i]    += je->rate[i];
 735                 dst->iops[i]    += je->iops[i];
 736         }
 737
 738         dst->elapsed_sec        += je->elapsed_sec;
 739
 740         if (je->eta_sec > dst->eta_sec)
 741                 dst->eta_sec = je->eta_sec;
 742 }
 743
 744 static void dec_jobs_eta(struct client_eta *eta)
 745 {
 746         if (!--eta->pending) {
 747                 display_thread_status(&eta->eta);
 748                 free(eta);
 749         }
 750 }
 751
 752 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 753 {
 754         struct fio_net_int_cmd *icmd = NULL;
 755         struct flist_head *entry;
 756
 757         flist_for_each(entry, &client->cmd_list) {
 758                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
 759
 760                 if (cmd->tag == (uintptr_t) icmd)
 761                         break;
 762
 763                 icmd = NULL;
 764         }
 765
 766         if (!icmd) {
 767                 log_err("fio: client: unable to find matching tag\n");
 768                 return;
 769         }
 770
 771         flist_del(&icmd->list);
 772         cmd->tag = icmd->saved_tag;
 773         free(icmd);
 774 }
 775
 776 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 777 {
 778         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 779         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 780
 781         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 782
 783         assert(client->eta_in_flight == eta);
 784
 785         client->eta_in_flight = NULL;
 786         flist_del_init(&client->eta_list);
 787
 788         convert_jobs_eta(je);
 789         sum_jobs_eta(&eta->eta, je);
 790         dec_jobs_eta(eta);
 791 }
 792
 793 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 794 {
 795         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 796         const char *os, *arch;
 797         char bit[16];
 798
 799         os = fio_get_os_string(probe->os);
 800         if (!os)
 801                 os = "unknown";
 802
 803         arch = fio_get_arch_string(probe->arch);
 804         if (!arch)
 805                 os = "unknown";
 806
 807         sprintf(bit, "%d-bit", probe->bpp * 8);
 808
 809         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
 810                 probe->hostname, probe->bigendian, bit, os, arch,
 811                 probe->fio_major, probe->fio_minor, probe->fio_patch);
 812
 813         if (!client->name)
 814                 client->name = strdup((char *) probe->hostname);
 815 }
 816
 817 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
 818 {
 819         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 820
 821         client->state = Client_started;
 822         client->jobs = le32_to_cpu(pdu->jobs);
 823 }
 824
 825 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
 826 {
 827         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
 828
 829         client->state = Client_stopped;
 830         client->error = le32_to_cpu(pdu->error);
 831
 832         if (client->error)
 833                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
 834 }
 835
 836 static int handle_client(struct fio_client *client)
 837 {
 838         struct fio_net_cmd *cmd;
 839
 840         dprint(FD_NET, "client: handle %s\n", client->hostname);
 841
 842         cmd = fio_net_recv_cmd(client->fd);
 843         if (!cmd)
 844                 return 0;
 845
 846         dprint(FD_NET, "client: got cmd op %s from %s\n",
 847                                 fio_server_op(cmd->opcode), client->hostname);
 848
 849         switch (cmd->opcode) {
 850         case FIO_NET_CMD_QUIT:
 851                 remove_client(client);
 852                 free(cmd);
 853                 break;
 854         case FIO_NET_CMD_TEXT: {
 855                 const char *buf = (const char *) cmd->payload;
 856                 const char *name;
 857                 int fio_unused ret;
 858
 859                 name = client->name ? client->name : client->hostname;
 860
 861                 if (!client->skip_newline)
 862                         fprintf(f_out, "<%s> ", name);
 863                 ret = fwrite(buf, cmd->pdu_len, 1, f_out);
 864                 fflush(f_out);
 865                 client->skip_newline = strchr(buf, '\n') == NULL;
 866                 free(cmd);
 867                 break;
 868                 }
 869         case FIO_NET_CMD_DU:
 870                 handle_du(client, cmd);
 871                 free(cmd);
 872                 break;
 873         case FIO_NET_CMD_TS:
 874                 handle_ts(cmd);
 875                 free(cmd);
 876                 break;
 877         case FIO_NET_CMD_GS:
 878                 handle_gs(cmd);
 879                 free(cmd);
 880                 break;
 881         case FIO_NET_CMD_ETA:
 882                 remove_reply_cmd(client, cmd);
 883                 handle_eta(client, cmd);
 884                 free(cmd);
 885                 break;
 886         case FIO_NET_CMD_PROBE:
 887                 remove_reply_cmd(client, cmd);
 888                 handle_probe(client, cmd);
 889                 free(cmd);
 890                 break;
 891         case FIO_NET_CMD_RUN:
 892                 client->state = Client_running;
 893                 free(cmd);
 894                 break;
 895         case FIO_NET_CMD_START:
 896                 handle_start(client, cmd);
 897                 free(cmd);
 898                 break;
 899         case FIO_NET_CMD_STOP:
 900                 handle_stop(client, cmd);
 901                 free(cmd);
 902                 break;
 903         default:
 904                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
 905                 free(cmd);
 906                 break;
 907         }
 908
 909         return 1;
 910 }
 911
 912 static void request_client_etas(void)
 913 {
 914         struct fio_client *client;
 915         struct flist_head *entry;
 916         struct client_eta *eta;
 917         int skipped = 0;
 918
 919         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
 920
 921         eta = malloc(sizeof(*eta));
 922         memset(&eta->eta, 0, sizeof(eta->eta));
 923         eta->pending = nr_clients;
 924
 925         flist_for_each(entry, &client_list) {
 926                 client = flist_entry(entry, struct fio_client, list);
 927
 928                 if (!flist_empty(&client->eta_list)) {
 929                         skipped++;
 930                         continue;
 931                 }
 932                 if (client->state != Client_running)
 933                         continue;
 934
 935                 assert(!client->eta_in_flight);
 936                 flist_add_tail(&client->eta_list, &eta_list);
 937                 client->eta_in_flight = eta;
 938                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
 939                                         (uintptr_t) eta, &client->cmd_list);
 940         }
 941
 942         while (skipped--)
 943                 dec_jobs_eta(eta);
 944
 945         dprint(FD_NET, "client: requested eta tag %p\n", eta);
 946 }
 947
 948 static int client_check_cmd_timeout(struct fio_client *client,
 949                                     struct timeval *now)
 950 {
 951         struct fio_net_int_cmd *cmd;
 952         struct flist_head *entry, *tmp;
 953         int ret = 0;
 954
 955         flist_for_each_safe(entry, tmp, &client->cmd_list) {
 956                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
 957
 958                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
 959                         continue;
 960
 961                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
 962                                                 fio_server_op(cmd->cmd.opcode));
 963                 flist_del(&cmd->list);
 964                 free(cmd);
 965                 ret = 1;
 966         }
 967
 968         return flist_empty(&client->cmd_list) && ret;
 969 }
 970
 971 static int fio_client_timed_out(void)
 972 {
 973         struct fio_client *client;
 974         struct flist_head *entry, *tmp;
 975         struct timeval tv;
 976         int ret = 0;
 977
 978         gettimeofday(&tv, NULL);
 979
 980         flist_for_each_safe(entry, tmp, &client_list) {
 981                 client = flist_entry(entry, struct fio_client, list);
 982
 983                 if (flist_empty(&client->cmd_list))
 984                         continue;
 985
 986                 if (!client_check_cmd_timeout(client, &tv))
 987                         continue;
 988
 989                 log_err("fio: client %s timed out\n", client->hostname);
 990                 remove_client(client);
 991                 ret = 1;
 992         }
 993
 994         return ret;
 995 }
 996
 997 int fio_handle_clients(void)
 998 {
 999         struct pollfd *pfds;
1000         int i, ret = 0, retval = 0;
1001
1002         gettimeofday(&eta_tv, NULL);
1003
1004         pfds = malloc(nr_clients * sizeof(struct pollfd));
1005
1006         sum_stat_clients = nr_clients;
1007         init_thread_stat(&client_ts);
1008         init_group_run_stat(&client_gs);
1009
1010         while (!exit_backend && nr_clients) {
1011                 struct flist_head *entry, *tmp;
1012                 struct fio_client *client;
1013
1014                 i = 0;
1015                 flist_for_each_safe(entry, tmp, &client_list) {
1016                         client = flist_entry(entry, struct fio_client, list);
1017
1018                         if (!client->sent_job &&
1019                             flist_empty(&client->cmd_list)) {
1020                                 remove_client(client);
1021                                 continue;
1022                         }
1023
1024                         pfds[i].fd = client->fd;
1025                         pfds[i].events = POLLIN;
1026                         i++;
1027                 }
1028
1029                 if (!nr_clients)
1030                         break;
1031
1032                 assert(i == nr_clients);
1033
1034                 do {
1035                         struct timeval tv;
1036
1037                         gettimeofday(&tv, NULL);
1038                         if (mtime_since(&eta_tv, &tv) >= 900) {
1039                                 request_client_etas();
1040                                 memcpy(&eta_tv, &tv, sizeof(tv));
1041
1042                                 if (fio_client_timed_out())
1043                                         break;
1044                         }
1045
1046                         ret = poll(pfds, nr_clients, 100);
1047                         if (ret < 0) {
1048                                 if (errno == EINTR)
1049                                         continue;
1050                                 log_err("fio: poll clients: %s\n", strerror(errno));
1051                                 break;
1052                         } else if (!ret)
1053                                 continue;
1054                 } while (ret <= 0);
1055
1056                 for (i = 0; i < nr_clients; i++) {
1057                         if (!(pfds[i].revents & POLLIN))
1058                                 continue;
1059
1060                         client = find_client_by_fd(pfds[i].fd);
1061                         if (!client) {
1062                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1063                                 continue;
1064                         }
1065                         if (!handle_client(client)) {
1066                                 log_info("client: host=%s disconnected\n",
1067                                                 client->hostname);
1068                                 remove_client(client);
1069                                 retval = 1;
1070                         } else if (client->error)
1071                                 retval = 1;
1072                         put_client(client);
1073                 }
1074         }
1075
1076         free(pfds);
1077         return retval;
1078 }