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         char **ini_file;
  63         unsigned int nr_ini_file;
  64 };
  65
  66 static struct timeval eta_tv;
  67
  68 enum {
  69         Client_created          = 0,
  70         Client_connected        = 1,
  71         Client_started          = 2,
  72         Client_running          = 3,
  73         Client_stopped          = 4,
  74         Client_exited           = 5,
  75 };
  76
  77 static FLIST_HEAD(client_list);
  78 static FLIST_HEAD(eta_list);
  79
  80 static FLIST_HEAD(arg_list);
  81
  82 static struct thread_stat client_ts;
  83 static struct group_run_stats client_gs;
  84 static int sum_stat_clients;
  85 static int sum_stat_nr;
  86
  87 #define FIO_CLIENT_HASH_BITS    7
  88 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  89 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  90 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  91
  92 static int handle_client(struct fio_client *client);
  93 static void dec_jobs_eta(struct client_eta *eta);
  94
  95 static void fio_client_add_hash(struct fio_client *client)
  96 {
  97         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  98
  99         bucket &= FIO_CLIENT_HASH_MASK;
 100         flist_add(&client->hash_list, &client_hash[bucket]);
 101 }
 102
 103 static void fio_client_remove_hash(struct fio_client *client)
 104 {
 105         if (!flist_empty(&client->hash_list))
 106                 flist_del_init(&client->hash_list);
 107 }
 108
 109 static void fio_init fio_client_hash_init(void)
 110 {
 111         int i;
 112
 113         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
 114                 INIT_FLIST_HEAD(&client_hash[i]);
 115 }
 116
 117 static struct fio_client *find_client_by_fd(int fd)
 118 {
 119         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
 120         struct fio_client *client;
 121         struct flist_head *entry;
 122
 123         flist_for_each(entry, &client_hash[bucket]) {
 124                 client = flist_entry(entry, struct fio_client, hash_list);
 125
 126                 if (client->fd == fd) {
 127                         client->refs++;
 128                         return client;
 129                 }
 130         }
 131
 132         return NULL;
 133 }
 134
 135 static void remove_client(struct fio_client *client)
 136 {
 137         assert(client->refs);
 138
 139         if (--client->refs)
 140                 return;
 141
 142         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 143         flist_del(&client->list);
 144
 145         fio_client_remove_hash(client);
 146
 147         if (!flist_empty(&client->eta_list)) {
 148                 flist_del_init(&client->eta_list);
 149                 dec_jobs_eta(client->eta_in_flight);
 150         }
 151
 152         free(client->hostname);
 153         if (client->argv)
 154                 free(client->argv);
 155         if (client->name)
 156                 free(client->name);
 157         while (client->nr_ini_file)
 158                 free(client->ini_file[--client->nr_ini_file]);
 159         if (client->ini_file)
 160                 free(client->ini_file);
 161
 162         free(client);
 163         nr_clients--;
 164         sum_stat_clients--;
 165 }
 166
 167 static void put_client(struct fio_client *client)
 168 {
 169         remove_client(client);
 170 }
 171
 172 static void __fio_client_add_cmd_option(struct fio_client *client,
 173                                         const char *opt)
 174 {
 175         int index;
 176
 177         index = client->argc++;
 178         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 179         client->argv[index] = strdup(opt);
 180         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 181 }
 182
 183 void fio_client_add_cmd_option(void *cookie, const char *opt)
 184 {
 185         struct fio_client *client = cookie;
 186         struct flist_head *entry;
 187
 188         if (!client || !opt)
 189                 return;
 190
 191         __fio_client_add_cmd_option(client, opt);
 192
 193         /*
 194          * Duplicate arguments to shared client group
 195          */
 196         flist_for_each(entry, &arg_list) {
 197                 client = flist_entry(entry, struct fio_client, arg_list);
 198
 199                 __fio_client_add_cmd_option(client, opt);
 200         }
 201 }
 202
 203 void fio_client_add_ini_file(void *cookie, const char *ini_file)
 204 {
 205         struct fio_client *client = cookie;
 206         size_t new_size;
 207
 208         dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file);
 209
 210         new_size = (client->nr_ini_file + 1) * sizeof(char *);
 211         client->ini_file = realloc(client->ini_file, new_size);
 212         client->ini_file[client->nr_ini_file] = strdup(ini_file);
 213         client->nr_ini_file++;
 214 }
 215
 216 int fio_client_add(const char *hostname, void **cookie)
 217 {
 218         struct fio_client *existing = *cookie;
 219         struct fio_client *client;
 220
 221         if (existing) {
 222                 /*
 223                  * We always add our "exec" name as the option, hence 1
 224                  * means empty.
 225                  */
 226                 if (existing->argc == 1)
 227                         flist_add_tail(&existing->arg_list, &arg_list);
 228                 else {
 229                         while (!flist_empty(&arg_list))
 230                                 flist_del_init(arg_list.next);
 231                 }
 232         }
 233
 234         client = malloc(sizeof(*client));
 235         memset(client, 0, sizeof(*client));
 236
 237         INIT_FLIST_HEAD(&client->list);
 238         INIT_FLIST_HEAD(&client->hash_list);
 239         INIT_FLIST_HEAD(&client->arg_list);
 240         INIT_FLIST_HEAD(&client->eta_list);
 241         INIT_FLIST_HEAD(&client->cmd_list);
 242
 243         if (fio_server_parse_string(hostname, &client->hostname,
 244                                         &client->is_sock, &client->port,
 245                                         &client->addr.sin_addr,
 246                                         &client->addr6.sin6_addr,
 247                                         &client->ipv6))
 248                 return -1;
 249
 250         client->fd = -1;
 251         client->refs = 1;
 252
 253         __fio_client_add_cmd_option(client, "fio");
 254
 255         flist_add(&client->list, &client_list);
 256         nr_clients++;
 257         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 258         *cookie = client;
 259         return 0;
 260 }
 261
 262 static int fio_client_connect_ip(struct fio_client *client)
 263 {
 264         struct sockaddr *addr;
 265         fio_socklen_t socklen;
 266         int fd, domain;
 267
 268         if (client->ipv6) {
 269                 client->addr6.sin6_family = AF_INET6;
 270                 client->addr6.sin6_port = htons(client->port);
 271                 domain = AF_INET6;
 272                 addr = (struct sockaddr *) &client->addr6;
 273                 socklen = sizeof(client->addr6);
 274         } else {
 275                 client->addr.sin_family = AF_INET;
 276                 client->addr.sin_port = htons(client->port);
 277                 domain = AF_INET;
 278                 addr = (struct sockaddr *) &client->addr;
 279                 socklen = sizeof(client->addr);
 280         }
 281
 282         fd = socket(domain, SOCK_STREAM, 0);
 283         if (fd < 0) {
 284                 log_err("fio: socket: %s\n", strerror(errno));
 285                 return -1;
 286         }
 287
 288         if (connect(fd, addr, socklen) < 0) {
 289                 log_err("fio: connect: %s\n", strerror(errno));
 290                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 291                                                                 client->port);
 292                 close(fd);
 293                 return -1;
 294         }
 295
 296         return fd;
 297 }
 298
 299 static int fio_client_connect_sock(struct fio_client *client)
 300 {
 301         struct sockaddr_un *addr = &client->addr_un;
 302         fio_socklen_t len;
 303         int fd;
 304
 305         memset(addr, 0, sizeof(*addr));
 306         addr->sun_family = AF_UNIX;
 307         strcpy(addr->sun_path, client->hostname);
 308
 309         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 310         if (fd < 0) {
 311                 log_err("fio: socket: %s\n", strerror(errno));
 312                 return -1;
 313         }
 314
 315         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 316         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 317                 log_err("fio: connect; %s\n", strerror(errno));
 318                 close(fd);
 319                 return -1;
 320         }
 321
 322         return fd;
 323 }
 324
 325 static int fio_client_connect(struct fio_client *client)
 326 {
 327         int fd;
 328
 329         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 330
 331         if (client->is_sock)
 332                 fd = fio_client_connect_sock(client);
 333         else
 334                 fd = fio_client_connect_ip(client);
 335
 336         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 337
 338         if (fd < 0)
 339                 return 1;
 340
 341         client->fd = fd;
 342         fio_client_add_hash(client);
 343         client->state = Client_connected;
 344         return 0;
 345 }
 346
 347 void fio_clients_terminate(void)
 348 {
 349         struct flist_head *entry;
 350         struct fio_client *client;
 351
 352         dprint(FD_NET, "client: terminate clients\n");
 353
 354         flist_for_each(entry, &client_list) {
 355                 client = flist_entry(entry, struct fio_client, list);
 356
 357                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0, NULL);
 358         }
 359 }
 360
 361 static void sig_int(int sig)
 362 {
 363         dprint(FD_NET, "client: got signal %d\n", sig);
 364         fio_clients_terminate();
 365 }
 366
 367 static void sig_show_status(int sig)
 368 {
 369         show_running_run_stats();
 370 }
 371
 372 static void client_signal_handler(void)
 373 {
 374         struct sigaction act;
 375
 376         memset(&act, 0, sizeof(act));
 377         act.sa_handler = sig_int;
 378         act.sa_flags = SA_RESTART;
 379         sigaction(SIGINT, &act, NULL);
 380
 381         memset(&act, 0, sizeof(act));
 382         act.sa_handler = sig_int;
 383         act.sa_flags = SA_RESTART;
 384         sigaction(SIGTERM, &act, NULL);
 385
 386 /* Windows uses SIGBREAK as a quit signal from other applications */
 387 #ifdef WIN32
 388         memset(&act, 0, sizeof(act));
 389         act.sa_handler = sig_int;
 390         act.sa_flags = SA_RESTART;
 391         sigaction(SIGBREAK, &act, NULL);
 392 #endif
 393
 394         memset(&act, 0, sizeof(act));
 395         act.sa_handler = sig_show_status;
 396         act.sa_flags = SA_RESTART;
 397         sigaction(SIGUSR1, &act, NULL);
 398 }
 399
 400 static void probe_client(struct fio_client *client)
 401 {
 402         dprint(FD_NET, "client: send probe\n");
 403
 404         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
 405 }
 406
 407 static int send_client_cmd_line(struct fio_client *client)
 408 {
 409         struct cmd_single_line_pdu *cslp;
 410         struct cmd_line_pdu *clp;
 411         unsigned long offset;
 412         unsigned int *lens;
 413         void *pdu;
 414         size_t mem;
 415         int i, ret;
 416
 417         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 418
 419         lens = malloc(client->argc * sizeof(unsigned int));
 420
 421         /*
 422          * Find out how much mem we need
 423          */
 424         for (i = 0, mem = 0; i < client->argc; i++) {
 425                 lens[i] = strlen(client->argv[i]) + 1;
 426                 mem += lens[i];
 427         }
 428
 429         /*
 430          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 431          */
 432         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 433
 434         pdu = malloc(mem);
 435         clp = pdu;
 436         offset = sizeof(*clp);
 437
 438         for (i = 0; i < client->argc; i++) {
 439                 uint16_t arg_len = lens[i];
 440
 441                 cslp = pdu + offset;
 442                 strcpy((char *) cslp->text, client->argv[i]);
 443                 cslp->len = cpu_to_le16(arg_len);
 444                 offset += sizeof(*cslp) + arg_len;
 445         }
 446
 447         free(lens);
 448         clp->lines = cpu_to_le16(client->argc);
 449         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
 450         free(pdu);
 451         return ret;
 452 }
 453
 454 int fio_clients_connect(void)
 455 {
 456         struct fio_client *client;
 457         struct flist_head *entry, *tmp;
 458         int ret;
 459
 460 #ifdef WIN32
 461         WSADATA wsd;
 462         WSAStartup(MAKEWORD(2,2), &wsd);
 463 #endif
 464
 465         dprint(FD_NET, "client: connect all\n");
 466
 467         client_signal_handler();
 468
 469         flist_for_each_safe(entry, tmp, &client_list) {
 470                 client = flist_entry(entry, struct fio_client, list);
 471
 472                 ret = fio_client_connect(client);
 473                 if (ret) {
 474                         remove_client(client);
 475                         continue;
 476                 }
 477
 478                 probe_client(client);
 479
 480                 if (client->argc > 1)
 481                         send_client_cmd_line(client);
 482         }
 483
 484         return !nr_clients;
 485 }
 486
 487 /*
 488  * Send file contents to server backend. We could use sendfile(), but to remain
 489  * more portable lets just read/write the darn thing.
 490  */
 491 static int fio_client_send_ini(struct fio_client *client, const char *filename)
 492 {
 493         struct stat sb;
 494         char *p, *buf;
 495         off_t len;
 496         int fd, ret;
 497
 498         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 499
 500         fd = open(filename, O_RDONLY);
 501         if (fd < 0) {
 502                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 503                 return 1;
 504         }
 505
 506         if (fstat(fd, &sb) < 0) {
 507                 log_err("fio: job file stat: %s\n", strerror(errno));
 508                 close(fd);
 509                 return 1;
 510         }
 511
 512         buf = malloc(sb.st_size);
 513
 514         len = sb.st_size;
 515         p = buf;
 516         do {
 517                 ret = read(fd, p, len);
 518                 if (ret > 0) {
 519                         len -= ret;
 520                         if (!len)
 521                                 break;
 522                         p += ret;
 523                         continue;
 524                 } else if (!ret)
 525                         break;
 526                 else if (errno == EAGAIN || errno == EINTR)
 527                         continue;
 528         } while (1);
 529
 530         if (len) {
 531                 log_err("fio: failed reading job file %s\n", filename);
 532                 close(fd);
 533                 free(buf);
 534                 return 1;
 535         }
 536
 537         client->sent_job = 1;
 538         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size, 0);
 539         free(buf);
 540         close(fd);
 541         return ret;
 542 }
 543
 544 int fio_clients_send_ini(const char *filename)
 545 {
 546         struct fio_client *client;
 547         struct flist_head *entry, *tmp;
 548
 549         flist_for_each_safe(entry, tmp, &client_list) {
 550                 client = flist_entry(entry, struct fio_client, list);
 551
 552                 if (client->nr_ini_file) {
 553                         int i;
 554
 555                         for (i = 0; i < client->nr_ini_file; i++) {
 556                                 const char *ini = client->ini_file[i];
 557
 558                                 if (fio_client_send_ini(client, ini)) {
 559                                         remove_client(client);
 560                                         break;
 561                                 }
 562                         }
 563                 } else if (!filename || fio_client_send_ini(client, filename))
 564                         remove_client(client);
 565
 566                 client->sent_job = 1;
 567         }
 568
 569         return !nr_clients;
 570 }
 571
 572 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 573 {
 574         dst->max_val    = le64_to_cpu(src->max_val);
 575         dst->min_val    = le64_to_cpu(src->min_val);
 576         dst->samples    = le64_to_cpu(src->samples);
 577
 578         /*
 579          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 580          */
 581         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 582         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 583 }
 584
 585 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 586 {
 587         int i, j;
 588
 589         dst->error      = le32_to_cpu(src->error);
 590         dst->groupid    = le32_to_cpu(src->groupid);
 591         dst->pid        = le32_to_cpu(src->pid);
 592         dst->members    = le32_to_cpu(src->members);
 593
 594         for (i = 0; i < 2; i++) {
 595                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 596                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 597                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 598                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 599         }
 600
 601         dst->usr_time           = le64_to_cpu(src->usr_time);
 602         dst->sys_time           = le64_to_cpu(src->sys_time);
 603         dst->ctx                = le64_to_cpu(src->ctx);
 604         dst->minf               = le64_to_cpu(src->minf);
 605         dst->majf               = le64_to_cpu(src->majf);
 606         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 607
 608         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 609                 fio_fp64_t *fps = &src->percentile_list[i];
 610                 fio_fp64_t *fpd = &dst->percentile_list[i];
 611
 612                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 613         }
 614
 615         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 616                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 617                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 618                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 619         }
 620
 621         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 622                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 623                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 624         }
 625
 626         for (i = 0; i < 2; i++)
 627                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 628                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 629
 630         for (i = 0; i < 3; i++) {
 631                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 632                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 633         }
 634
 635         dst->total_submit       = le64_to_cpu(src->total_submit);
 636         dst->total_complete     = le64_to_cpu(src->total_complete);
 637
 638         for (i = 0; i < 2; i++) {
 639                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 640                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 641         }
 642
 643         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 644         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 645         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 646         dst->first_error        = le32_to_cpu(src->first_error);
 647         dst->kb_base            = le32_to_cpu(src->kb_base);
 648 }
 649
 650 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 651 {
 652         int i;
 653
 654         for (i = 0; i < 2; i++) {
 655                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 656                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 657                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 658                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 659                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 660                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 661         }
 662
 663         dst->kb_base    = le32_to_cpu(src->kb_base);
 664         dst->groupid    = le32_to_cpu(src->groupid);
 665 }
 666
 667 static void handle_ts(struct fio_net_cmd *cmd)
 668 {
 669         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 670
 671         convert_ts(&p->ts, &p->ts);
 672         convert_gs(&p->rs, &p->rs);
 673
 674         show_thread_status(&p->ts, &p->rs);
 675
 676         if (sum_stat_clients == 1)
 677                 return;
 678
 679         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 680         sum_group_stats(&client_gs, &p->rs);
 681
 682         client_ts.members++;
 683         client_ts.groupid = p->ts.groupid;
 684
 685         if (++sum_stat_nr == sum_stat_clients) {
 686                 strcpy(client_ts.name, "All clients");
 687                 show_thread_status(&client_ts, &client_gs);
 688         }
 689 }
 690
 691 static void handle_gs(struct fio_net_cmd *cmd)
 692 {
 693         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 694
 695         convert_gs(gs, gs);
 696         show_group_stats(gs);
 697 }
 698
 699 static void convert_agg(struct disk_util_agg *agg)
 700 {
 701         int i;
 702
 703         for (i = 0; i < 2; i++) {
 704                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 705                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 706                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 707                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 708         }
 709
 710         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 711         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 712         agg->slavecount         = le32_to_cpu(agg->slavecount);
 713         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 714 }
 715
 716 static void convert_dus(struct disk_util_stat *dus)
 717 {
 718         int i;
 719
 720         for (i = 0; i < 2; i++) {
 721                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
 722                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
 723                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
 724                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
 725         }
 726
 727         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
 728         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
 729         dus->msec               = le64_to_cpu(dus->msec);
 730 }
 731
 732 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 733 {
 734         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 735
 736         convert_dus(&du->dus);
 737         convert_agg(&du->agg);
 738
 739         if (!client->disk_stats_shown) {
 740                 client->disk_stats_shown = 1;
 741                 log_info("\nDisk stats (read/write):\n");
 742         }
 743
 744         print_disk_util(&du->dus, &du->agg, output_format == FIO_OUTPUT_TERSE);
 745 }
 746
 747 static void convert_jobs_eta(struct jobs_eta *je)
 748 {
 749         int i;
 750
 751         je->nr_running          = le32_to_cpu(je->nr_running);
 752         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 753         je->nr_pending          = le32_to_cpu(je->nr_pending);
 754         je->files_open          = le32_to_cpu(je->files_open);
 755         je->m_rate              = le32_to_cpu(je->m_rate);
 756         je->t_rate              = le32_to_cpu(je->t_rate);
 757         je->m_iops              = le32_to_cpu(je->m_iops);
 758         je->t_iops              = le32_to_cpu(je->t_iops);
 759
 760         for (i = 0; i < 2; i++) {
 761                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 762                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 763         }
 764
 765         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 766         je->eta_sec             = le64_to_cpu(je->eta_sec);
 767         je->is_pow2             = le32_to_cpu(je->is_pow2);
 768 }
 769
 770 static void sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 771 {
 772         int i;
 773
 774         dst->nr_running         += je->nr_running;
 775         dst->nr_ramp            += je->nr_ramp;
 776         dst->nr_pending         += je->nr_pending;
 777         dst->files_open         += je->files_open;
 778         dst->m_rate             += je->m_rate;
 779         dst->t_rate             += je->t_rate;
 780         dst->m_iops             += je->m_iops;
 781         dst->t_iops             += je->t_iops;
 782
 783         for (i = 0; i < 2; i++) {
 784                 dst->rate[i]    += je->rate[i];
 785                 dst->iops[i]    += je->iops[i];
 786         }
 787
 788         dst->elapsed_sec        += je->elapsed_sec;
 789
 790         if (je->eta_sec > dst->eta_sec)
 791                 dst->eta_sec = je->eta_sec;
 792 }
 793
 794 static void dec_jobs_eta(struct client_eta *eta)
 795 {
 796         if (!--eta->pending) {
 797                 display_thread_status(&eta->eta);
 798                 free(eta);
 799         }
 800 }
 801
 802 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 803 {
 804         struct fio_net_int_cmd *icmd = NULL;
 805         struct flist_head *entry;
 806
 807         flist_for_each(entry, &client->cmd_list) {
 808                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
 809
 810                 if (cmd->tag == (uintptr_t) icmd)
 811                         break;
 812
 813                 icmd = NULL;
 814         }
 815
 816         if (!icmd) {
 817                 log_err("fio: client: unable to find matching tag\n");
 818                 return;
 819         }
 820
 821         flist_del(&icmd->list);
 822         cmd->tag = icmd->saved_tag;
 823         free(icmd);
 824 }
 825
 826 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 827 {
 828         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 829         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 830
 831         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 832
 833         assert(client->eta_in_flight == eta);
 834
 835         client->eta_in_flight = NULL;
 836         flist_del_init(&client->eta_list);
 837
 838         convert_jobs_eta(je);
 839         sum_jobs_eta(&eta->eta, je);
 840         dec_jobs_eta(eta);
 841 }
 842
 843 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 844 {
 845         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 846         const char *os, *arch;
 847         char bit[16];
 848
 849         os = fio_get_os_string(probe->os);
 850         if (!os)
 851                 os = "unknown";
 852
 853         arch = fio_get_arch_string(probe->arch);
 854         if (!arch)
 855                 os = "unknown";
 856
 857         sprintf(bit, "%d-bit", probe->bpp * 8);
 858
 859         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s\n",
 860                 probe->hostname, probe->bigendian, bit, os, arch,
 861                 probe->fio_version);
 862
 863         if (!client->name)
 864                 client->name = strdup((char *) probe->hostname);
 865 }
 866
 867 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
 868 {
 869         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 870
 871         client->state = Client_started;
 872         client->jobs = le32_to_cpu(pdu->jobs);
 873 }
 874
 875 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
 876 {
 877         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
 878
 879         client->state = Client_stopped;
 880         client->error = le32_to_cpu(pdu->error);
 881
 882         if (client->error)
 883                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
 884 }
 885
 886 static int handle_client(struct fio_client *client)
 887 {
 888         struct fio_net_cmd *cmd;
 889
 890         dprint(FD_NET, "client: handle %s\n", client->hostname);
 891
 892         cmd = fio_net_recv_cmd(client->fd);
 893         if (!cmd)
 894                 return 0;
 895
 896         dprint(FD_NET, "client: got cmd op %s from %s\n",
 897                                 fio_server_op(cmd->opcode), client->hostname);
 898
 899         switch (cmd->opcode) {
 900         case FIO_NET_CMD_QUIT:
 901                 remove_client(client);
 902                 free(cmd);
 903                 break;
 904         case FIO_NET_CMD_TEXT: {
 905                 const char *buf = (const char *) cmd->payload;
 906                 const char *name;
 907                 int fio_unused ret;
 908
 909                 name = client->name ? client->name : client->hostname;
 910
 911                 if (!client->skip_newline)
 912                         fprintf(f_out, "<%s> ", name);
 913                 ret = fwrite(buf, cmd->pdu_len, 1, f_out);
 914                 fflush(f_out);
 915                 client->skip_newline = strchr(buf, '\n') == NULL;
 916                 free(cmd);
 917                 break;
 918                 }
 919         case FIO_NET_CMD_DU:
 920                 handle_du(client, cmd);
 921                 free(cmd);
 922                 break;
 923         case FIO_NET_CMD_TS:
 924                 handle_ts(cmd);
 925                 free(cmd);
 926                 break;
 927         case FIO_NET_CMD_GS:
 928                 handle_gs(cmd);
 929                 free(cmd);
 930                 break;
 931         case FIO_NET_CMD_ETA:
 932                 remove_reply_cmd(client, cmd);
 933                 handle_eta(client, cmd);
 934                 free(cmd);
 935                 break;
 936         case FIO_NET_CMD_PROBE:
 937                 remove_reply_cmd(client, cmd);
 938                 handle_probe(client, cmd);
 939                 free(cmd);
 940                 break;
 941         case FIO_NET_CMD_RUN:
 942                 client->state = Client_running;
 943                 free(cmd);
 944                 break;
 945         case FIO_NET_CMD_START:
 946                 handle_start(client, cmd);
 947                 free(cmd);
 948                 break;
 949         case FIO_NET_CMD_STOP:
 950                 handle_stop(client, cmd);
 951                 free(cmd);
 952                 break;
 953         default:
 954                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
 955                 free(cmd);
 956                 break;
 957         }
 958
 959         return 1;
 960 }
 961
 962 static void request_client_etas(void)
 963 {
 964         struct fio_client *client;
 965         struct flist_head *entry;
 966         struct client_eta *eta;
 967         int skipped = 0;
 968
 969         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
 970
 971         eta = malloc(sizeof(*eta));
 972         memset(&eta->eta, 0, sizeof(eta->eta));
 973         eta->pending = nr_clients;
 974
 975         flist_for_each(entry, &client_list) {
 976                 client = flist_entry(entry, struct fio_client, list);
 977
 978                 if (!flist_empty(&client->eta_list)) {
 979                         skipped++;
 980                         continue;
 981                 }
 982                 if (client->state != Client_running)
 983                         continue;
 984
 985                 assert(!client->eta_in_flight);
 986                 flist_add_tail(&client->eta_list, &eta_list);
 987                 client->eta_in_flight = eta;
 988                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
 989                                         (uintptr_t) eta, &client->cmd_list);
 990         }
 991
 992         while (skipped--)
 993                 dec_jobs_eta(eta);
 994
 995         dprint(FD_NET, "client: requested eta tag %p\n", eta);
 996 }
 997
 998 static int client_check_cmd_timeout(struct fio_client *client,
 999                                     struct timeval *now)
1000 {
1001         struct fio_net_int_cmd *cmd;
1002         struct flist_head *entry, *tmp;
1003         int ret = 0;
1004
1005         flist_for_each_safe(entry, tmp, &client->cmd_list) {
1006                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
1007
1008                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
1009                         continue;
1010
1011                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1012                                                 fio_server_op(cmd->cmd.opcode));
1013                 flist_del(&cmd->list);
1014                 free(cmd);
1015                 ret = 1;
1016         }
1017
1018         return flist_empty(&client->cmd_list) && ret;
1019 }
1020
1021 static int fio_client_timed_out(void)
1022 {
1023         struct fio_client *client;
1024         struct flist_head *entry, *tmp;
1025         struct timeval tv;
1026         int ret = 0;
1027
1028         gettimeofday(&tv, NULL);
1029
1030         flist_for_each_safe(entry, tmp, &client_list) {
1031                 client = flist_entry(entry, struct fio_client, list);
1032
1033                 if (flist_empty(&client->cmd_list))
1034                         continue;
1035
1036                 if (!client_check_cmd_timeout(client, &tv))
1037                         continue;
1038
1039                 log_err("fio: client %s timed out\n", client->hostname);
1040                 remove_client(client);
1041                 ret = 1;
1042         }
1043
1044         return ret;
1045 }
1046
1047 int fio_handle_clients(void)
1048 {
1049         struct pollfd *pfds;
1050         int i, ret = 0, retval = 0;
1051
1052         gettimeofday(&eta_tv, NULL);
1053
1054         pfds = malloc(nr_clients * sizeof(struct pollfd));
1055
1056         sum_stat_clients = nr_clients;
1057         init_thread_stat(&client_ts);
1058         init_group_run_stat(&client_gs);
1059
1060         while (!exit_backend && nr_clients) {
1061                 struct flist_head *entry, *tmp;
1062                 struct fio_client *client;
1063
1064                 i = 0;
1065                 flist_for_each_safe(entry, tmp, &client_list) {
1066                         client = flist_entry(entry, struct fio_client, list);
1067
1068                         if (!client->sent_job &&
1069                             flist_empty(&client->cmd_list)) {
1070                                 remove_client(client);
1071                                 continue;
1072                         }
1073
1074                         pfds[i].fd = client->fd;
1075                         pfds[i].events = POLLIN;
1076                         i++;
1077                 }
1078
1079                 if (!nr_clients)
1080                         break;
1081
1082                 assert(i == nr_clients);
1083
1084                 do {
1085                         struct timeval tv;
1086
1087                         gettimeofday(&tv, NULL);
1088                         if (mtime_since(&eta_tv, &tv) >= 900) {
1089                                 request_client_etas();
1090                                 memcpy(&eta_tv, &tv, sizeof(tv));
1091
1092                                 if (fio_client_timed_out())
1093                                         break;
1094                         }
1095
1096                         ret = poll(pfds, nr_clients, 100);
1097                         if (ret < 0) {
1098                                 if (errno == EINTR)
1099                                         continue;
1100                                 log_err("fio: poll clients: %s\n", strerror(errno));
1101                                 break;
1102                         } else if (!ret)
1103                                 continue;
1104                 } while (ret <= 0);
1105
1106                 for (i = 0; i < nr_clients; i++) {
1107                         if (!(pfds[i].revents & POLLIN))
1108                                 continue;
1109
1110                         client = find_client_by_fd(pfds[i].fd);
1111                         if (!client) {
1112                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1113                                 continue;
1114                         }
1115                         if (!handle_client(client)) {
1116                                 log_info("client: host=%s disconnected\n",
1117                                                 client->hostname);
1118                                 remove_client(client);
1119                                 retval = 1;
1120                         } else if (client->error)
1121                                 retval = 1;
1122                         put_client(client);
1123                 }
1124         }
1125
1126         free(pfds);
1127         return retval;
1128 }