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