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         unsigned int pending;
  25         struct jobs_eta eta;
  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         unsigned int nr_stat;
  51         int error;
  52         int ipv6;
  53         int sent_job;
  54         int did_stat;
  55
  56         struct flist_head eta_list;
  57         struct client_eta *eta_in_flight;
  58
  59         struct flist_head cmd_list;
  60
  61         uint16_t argc;
  62         char **argv;
  63
  64         char **ini_file;
  65         unsigned int nr_ini_file;
  66 };
  67
  68 static struct timeval eta_tv;
  69
  70 enum {
  71         Client_created          = 0,
  72         Client_connected        = 1,
  73         Client_started          = 2,
  74         Client_running          = 3,
  75         Client_stopped          = 4,
  76         Client_exited           = 5,
  77 };
  78
  79 static FLIST_HEAD(client_list);
  80 static FLIST_HEAD(eta_list);
  81
  82 static FLIST_HEAD(arg_list);
  83
  84 static struct thread_stat client_ts;
  85 static struct group_run_stats client_gs;
  86 static int sum_stat_clients = 0;
  87 static int sum_stat_nr;
  88 static int do_output_all_clients;
  89
  90 #define FIO_CLIENT_HASH_BITS    7
  91 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  92 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  93 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  94
  95 static int handle_client(struct fio_client *client);
  96 static void dec_jobs_eta(struct client_eta *eta);
  97
  98 static void fio_client_add_hash(struct fio_client *client)
  99 {
 100         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
 101
 102         bucket &= FIO_CLIENT_HASH_MASK;
 103         flist_add(&client->hash_list, &client_hash[bucket]);
 104 }
 105
 106 static void fio_client_remove_hash(struct fio_client *client)
 107 {
 108         if (!flist_empty(&client->hash_list))
 109                 flist_del_init(&client->hash_list);
 110 }
 111
 112 static void fio_init fio_client_hash_init(void)
 113 {
 114         int i;
 115
 116         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
 117                 INIT_FLIST_HEAD(&client_hash[i]);
 118 }
 119
 120 static struct fio_client *find_client_by_fd(int fd)
 121 {
 122         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
 123         struct fio_client *client;
 124         struct flist_head *entry;
 125
 126         flist_for_each(entry, &client_hash[bucket]) {
 127                 client = flist_entry(entry, struct fio_client, hash_list);
 128
 129                 if (client->fd == fd) {
 130                         client->refs++;
 131                         return client;
 132                 }
 133         }
 134
 135         return NULL;
 136 }
 137
 138 static void remove_client(struct fio_client *client)
 139 {
 140         assert(client->refs);
 141
 142         if (--client->refs)
 143                 return;
 144
 145         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 146         flist_del(&client->list);
 147
 148         fio_client_remove_hash(client);
 149
 150         if (!flist_empty(&client->eta_list)) {
 151                 flist_del_init(&client->eta_list);
 152                 dec_jobs_eta(client->eta_in_flight);
 153         }
 154
 155         free(client->hostname);
 156         if (client->argv)
 157                 free(client->argv);
 158         if (client->name)
 159                 free(client->name);
 160         while (client->nr_ini_file)
 161                 free(client->ini_file[--client->nr_ini_file]);
 162         if (client->ini_file)
 163                 free(client->ini_file);
 164
 165         if (!client->did_stat)
 166                 sum_stat_clients -= client->nr_stat;
 167
 168         free(client);
 169         nr_clients--;
 170 }
 171
 172 static void put_client(struct fio_client *client)
 173 {
 174         remove_client(client);
 175 }
 176
 177 static void __fio_client_add_cmd_option(struct fio_client *client,
 178                                         const char *opt)
 179 {
 180         int index;
 181
 182         index = client->argc++;
 183         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 184         client->argv[index] = strdup(opt);
 185         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 186 }
 187
 188 void fio_client_add_cmd_option(void *cookie, const char *opt)
 189 {
 190         struct fio_client *client = cookie;
 191         struct flist_head *entry;
 192
 193         if (!client || !opt)
 194                 return;
 195
 196         __fio_client_add_cmd_option(client, opt);
 197
 198         /*
 199          * Duplicate arguments to shared client group
 200          */
 201         flist_for_each(entry, &arg_list) {
 202                 client = flist_entry(entry, struct fio_client, arg_list);
 203
 204                 __fio_client_add_cmd_option(client, opt);
 205         }
 206 }
 207
 208 void fio_client_add_ini_file(void *cookie, const char *ini_file)
 209 {
 210         struct fio_client *client = cookie;
 211         size_t new_size;
 212
 213         dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file);
 214
 215         new_size = (client->nr_ini_file + 1) * sizeof(char *);
 216         client->ini_file = realloc(client->ini_file, new_size);
 217         client->ini_file[client->nr_ini_file] = strdup(ini_file);
 218         client->nr_ini_file++;
 219 }
 220
 221 int fio_client_add(const char *hostname, void **cookie)
 222 {
 223         struct fio_client *existing = *cookie;
 224         struct fio_client *client;
 225
 226         if (existing) {
 227                 /*
 228                  * We always add our "exec" name as the option, hence 1
 229                  * means empty.
 230                  */
 231                 if (existing->argc == 1)
 232                         flist_add_tail(&existing->arg_list, &arg_list);
 233                 else {
 234                         while (!flist_empty(&arg_list))
 235                                 flist_del_init(arg_list.next);
 236                 }
 237         }
 238
 239         client = malloc(sizeof(*client));
 240         memset(client, 0, sizeof(*client));
 241
 242         INIT_FLIST_HEAD(&client->list);
 243         INIT_FLIST_HEAD(&client->hash_list);
 244         INIT_FLIST_HEAD(&client->arg_list);
 245         INIT_FLIST_HEAD(&client->eta_list);
 246         INIT_FLIST_HEAD(&client->cmd_list);
 247
 248         if (fio_server_parse_string(hostname, &client->hostname,
 249                                         &client->is_sock, &client->port,
 250                                         &client->addr.sin_addr,
 251                                         &client->addr6.sin6_addr,
 252                                         &client->ipv6))
 253                 return -1;
 254
 255         client->fd = -1;
 256         client->refs = 1;
 257
 258         __fio_client_add_cmd_option(client, "fio");
 259
 260         flist_add(&client->list, &client_list);
 261         nr_clients++;
 262         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 263         *cookie = client;
 264         return 0;
 265 }
 266
 267 static int fio_client_connect_ip(struct fio_client *client)
 268 {
 269         struct sockaddr *addr;
 270         socklen_t socklen;
 271         int fd, domain;
 272
 273         if (client->ipv6) {
 274                 client->addr6.sin6_family = AF_INET6;
 275                 client->addr6.sin6_port = htons(client->port);
 276                 domain = AF_INET6;
 277                 addr = (struct sockaddr *) &client->addr6;
 278                 socklen = sizeof(client->addr6);
 279         } else {
 280                 client->addr.sin_family = AF_INET;
 281                 client->addr.sin_port = htons(client->port);
 282                 domain = AF_INET;
 283                 addr = (struct sockaddr *) &client->addr;
 284                 socklen = sizeof(client->addr);
 285         }
 286
 287         fd = socket(domain, SOCK_STREAM, 0);
 288         if (fd < 0) {
 289                 log_err("fio: socket: %s\n", strerror(errno));
 290                 return -1;
 291         }
 292
 293         if (connect(fd, addr, socklen) < 0) {
 294                 log_err("fio: connect: %s\n", strerror(errno));
 295                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 296                                                                 client->port);
 297                 close(fd);
 298                 return -1;
 299         }
 300
 301         return fd;
 302 }
 303
 304 static int fio_client_connect_sock(struct fio_client *client)
 305 {
 306         struct sockaddr_un *addr = &client->addr_un;
 307         socklen_t len;
 308         int fd;
 309
 310         memset(addr, 0, sizeof(*addr));
 311         addr->sun_family = AF_UNIX;
 312         strcpy(addr->sun_path, client->hostname);
 313
 314         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 315         if (fd < 0) {
 316                 log_err("fio: socket: %s\n", strerror(errno));
 317                 return -1;
 318         }
 319
 320         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 321         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 322                 log_err("fio: connect; %s\n", strerror(errno));
 323                 close(fd);
 324                 return -1;
 325         }
 326
 327         return fd;
 328 }
 329
 330 static int fio_client_connect(struct fio_client *client)
 331 {
 332         int fd;
 333
 334         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 335
 336         if (client->is_sock)
 337                 fd = fio_client_connect_sock(client);
 338         else
 339                 fd = fio_client_connect_ip(client);
 340
 341         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 342
 343         if (fd < 0)
 344                 return 1;
 345
 346         client->fd = fd;
 347         fio_client_add_hash(client);
 348         client->state = Client_connected;
 349         return 0;
 350 }
 351
 352 void fio_clients_terminate(void)
 353 {
 354         struct flist_head *entry;
 355         struct fio_client *client;
 356
 357         dprint(FD_NET, "client: terminate clients\n");
 358
 359         flist_for_each(entry, &client_list) {
 360                 client = flist_entry(entry, struct fio_client, list);
 361
 362                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0, NULL);
 363         }
 364 }
 365
 366 static void sig_int(int sig)
 367 {
 368         dprint(FD_NET, "client: got signal %d\n", sig);
 369         fio_clients_terminate();
 370 }
 371
 372 static void sig_show_status(int sig)
 373 {
 374         show_running_run_stats();
 375 }
 376
 377 static void client_signal_handler(void)
 378 {
 379         struct sigaction act;
 380
 381         memset(&act, 0, sizeof(act));
 382         act.sa_handler = sig_int;
 383         act.sa_flags = SA_RESTART;
 384         sigaction(SIGINT, &act, NULL);
 385
 386         memset(&act, 0, sizeof(act));
 387         act.sa_handler = sig_int;
 388         act.sa_flags = SA_RESTART;
 389         sigaction(SIGTERM, &act, NULL);
 390
 391 /* Windows uses SIGBREAK as a quit signal from other applications */
 392 #ifdef WIN32
 393         memset(&act, 0, sizeof(act));
 394         act.sa_handler = sig_int;
 395         act.sa_flags = SA_RESTART;
 396         sigaction(SIGBREAK, &act, NULL);
 397 #endif
 398
 399         memset(&act, 0, sizeof(act));
 400         act.sa_handler = sig_show_status;
 401         act.sa_flags = SA_RESTART;
 402         sigaction(SIGUSR1, &act, NULL);
 403 }
 404
 405 static void probe_client(struct fio_client *client)
 406 {
 407         dprint(FD_NET, "client: send probe\n");
 408
 409         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
 410 }
 411
 412 static int send_client_cmd_line(struct fio_client *client)
 413 {
 414         struct cmd_single_line_pdu *cslp;
 415         struct cmd_line_pdu *clp;
 416         unsigned long offset;
 417         unsigned int *lens;
 418         void *pdu;
 419         size_t mem;
 420         int i, ret;
 421
 422         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 423
 424         lens = malloc(client->argc * sizeof(unsigned int));
 425
 426         /*
 427          * Find out how much mem we need
 428          */
 429         for (i = 0, mem = 0; i < client->argc; i++) {
 430                 lens[i] = strlen(client->argv[i]) + 1;
 431                 mem += lens[i];
 432         }
 433
 434         /*
 435          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 436          */
 437         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 438
 439         pdu = malloc(mem);
 440         clp = pdu;
 441         offset = sizeof(*clp);
 442
 443         for (i = 0; i < client->argc; i++) {
 444                 uint16_t arg_len = lens[i];
 445
 446                 cslp = pdu + offset;
 447                 strcpy((char *) cslp->text, client->argv[i]);
 448                 cslp->len = cpu_to_le16(arg_len);
 449                 offset += sizeof(*cslp) + arg_len;
 450         }
 451
 452         free(lens);
 453         clp->lines = cpu_to_le16(client->argc);
 454         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
 455         free(pdu);
 456         return ret;
 457 }
 458
 459 int fio_clients_connect(void)
 460 {
 461         struct fio_client *client;
 462         struct flist_head *entry, *tmp;
 463         int ret;
 464
 465 #ifdef WIN32
 466         WSADATA wsd;
 467         WSAStartup(MAKEWORD(2,2), &wsd);
 468 #endif
 469
 470         dprint(FD_NET, "client: connect all\n");
 471
 472         client_signal_handler();
 473
 474         flist_for_each_safe(entry, tmp, &client_list) {
 475                 client = flist_entry(entry, struct fio_client, list);
 476
 477                 ret = fio_client_connect(client);
 478                 if (ret) {
 479                         remove_client(client);
 480                         continue;
 481                 }
 482
 483                 probe_client(client);
 484
 485                 if (client->argc > 1)
 486                         send_client_cmd_line(client);
 487         }
 488
 489         return !nr_clients;
 490 }
 491
 492 /*
 493  * Send file contents to server backend. We could use sendfile(), but to remain
 494  * more portable lets just read/write the darn thing.
 495  */
 496 static int fio_client_send_ini(struct fio_client *client, const char *filename)
 497 {
 498         struct stat sb;
 499         char *p, *buf;
 500         off_t len;
 501         int fd, ret;
 502
 503         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 504
 505         fd = open(filename, O_RDONLY);
 506         if (fd < 0) {
 507                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 508                 return 1;
 509         }
 510
 511         if (fstat(fd, &sb) < 0) {
 512                 log_err("fio: job file stat: %s\n", strerror(errno));
 513                 close(fd);
 514                 return 1;
 515         }
 516
 517         buf = malloc(sb.st_size);
 518
 519         len = sb.st_size;
 520         p = buf;
 521         do {
 522                 ret = read(fd, p, len);
 523                 if (ret > 0) {
 524                         len -= ret;
 525                         if (!len)
 526                                 break;
 527                         p += ret;
 528                         continue;
 529                 } else if (!ret)
 530                         break;
 531                 else if (errno == EAGAIN || errno == EINTR)
 532                         continue;
 533         } while (1);
 534
 535         if (len) {
 536                 log_err("fio: failed reading job file %s\n", filename);
 537                 close(fd);
 538                 free(buf);
 539                 return 1;
 540         }
 541
 542         client->sent_job = 1;
 543         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size, 0);
 544         free(buf);
 545         close(fd);
 546         return ret;
 547 }
 548
 549 int fio_clients_send_ini(const char *filename)
 550 {
 551         struct fio_client *client;
 552         struct flist_head *entry, *tmp;
 553
 554         flist_for_each_safe(entry, tmp, &client_list) {
 555                 client = flist_entry(entry, struct fio_client, list);
 556
 557                 if (client->nr_ini_file) {
 558                         int i;
 559
 560                         for (i = 0; i < client->nr_ini_file; i++) {
 561                                 const char *ini = client->ini_file[i];
 562
 563                                 if (fio_client_send_ini(client, ini)) {
 564                                         remove_client(client);
 565                                         break;
 566                                 }
 567                         }
 568                 } else if (!filename || fio_client_send_ini(client, filename))
 569                         remove_client(client);
 570
 571                 client->sent_job = 1;
 572         }
 573
 574         return !nr_clients;
 575 }
 576
 577 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 578 {
 579         dst->max_val    = le64_to_cpu(src->max_val);
 580         dst->min_val    = le64_to_cpu(src->min_val);
 581         dst->samples    = le64_to_cpu(src->samples);
 582
 583         /*
 584          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 585          */
 586         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 587         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 588 }
 589
 590 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 591 {
 592         int i, j;
 593
 594         dst->error      = le32_to_cpu(src->error);
 595         dst->groupid    = le32_to_cpu(src->groupid);
 596         dst->pid        = le32_to_cpu(src->pid);
 597         dst->members    = le32_to_cpu(src->members);
 598         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
 599
 600         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 601                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 602                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 603                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 604                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 605         }
 606
 607         dst->usr_time           = le64_to_cpu(src->usr_time);
 608         dst->sys_time           = le64_to_cpu(src->sys_time);
 609         dst->ctx                = le64_to_cpu(src->ctx);
 610         dst->minf               = le64_to_cpu(src->minf);
 611         dst->majf               = le64_to_cpu(src->majf);
 612         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 613
 614         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 615                 fio_fp64_t *fps = &src->percentile_list[i];
 616                 fio_fp64_t *fpd = &dst->percentile_list[i];
 617
 618                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 619         }
 620
 621         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 622                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 623                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 624                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 625         }
 626
 627         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 628                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 629                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 630         }
 631
 632         for (i = 0; i < DDIR_RWDIR_CNT; i++)
 633                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 634                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 635
 636         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 637                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 638                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 639         }
 640
 641         dst->total_submit       = le64_to_cpu(src->total_submit);
 642         dst->total_complete     = le64_to_cpu(src->total_complete);
 643
 644         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 645                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 646                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 647         }
 648
 649         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 650         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 651         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 652         dst->first_error        = le32_to_cpu(src->first_error);
 653         dst->kb_base            = le32_to_cpu(src->kb_base);
 654 }
 655
 656 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 657 {
 658         int i;
 659
 660         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 661                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 662                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 663                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 664                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 665                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 666                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 667         }
 668
 669         dst->kb_base    = le32_to_cpu(src->kb_base);
 670         dst->groupid    = le32_to_cpu(src->groupid);
 671         dst->unified_rw_rep     = le32_to_cpu(src->unified_rw_rep);
 672 }
 673
 674 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
 675 {
 676         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 677
 678         convert_ts(&p->ts, &p->ts);
 679         convert_gs(&p->rs, &p->rs);
 680
 681         show_thread_status(&p->ts, &p->rs);
 682         client->did_stat = 1;
 683
 684         if (!do_output_all_clients)
 685                 return;
 686
 687         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 688         sum_group_stats(&client_gs, &p->rs);
 689
 690         client_ts.members++;
 691         client_ts.groupid = p->ts.groupid;
 692         client_ts.unified_rw_rep = p->ts.unified_rw_rep;
 693
 694         if (++sum_stat_nr == sum_stat_clients) {
 695                 strcpy(client_ts.name, "All clients");
 696                 show_thread_status(&client_ts, &client_gs);
 697         }
 698 }
 699
 700 static void handle_gs(struct fio_net_cmd *cmd)
 701 {
 702         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 703
 704         convert_gs(gs, gs);
 705         show_group_stats(gs);
 706 }
 707
 708 static void convert_agg(struct disk_util_agg *agg)
 709 {
 710         int i;
 711
 712         for (i = 0; i < 2; i++) {
 713                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 714                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 715                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 716                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 717         }
 718
 719         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 720         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 721         agg->slavecount         = le32_to_cpu(agg->slavecount);
 722         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 723 }
 724
 725 static void convert_dus(struct disk_util_stat *dus)
 726 {
 727         int i;
 728
 729         for (i = 0; i < 2; i++) {
 730                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
 731                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
 732                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
 733                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
 734         }
 735
 736         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
 737         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
 738         dus->msec               = le64_to_cpu(dus->msec);
 739 }
 740
 741 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 742 {
 743         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 744
 745         convert_dus(&du->dus);
 746         convert_agg(&du->agg);
 747
 748         if (!client->disk_stats_shown) {
 749                 client->disk_stats_shown = 1;
 750                 log_info("\nDisk stats (read/write):\n");
 751         }
 752
 753         print_disk_util(&du->dus, &du->agg, output_format == FIO_OUTPUT_TERSE);
 754 }
 755
 756 static void convert_jobs_eta(struct jobs_eta *je)
 757 {
 758         int i;
 759
 760         je->nr_running          = le32_to_cpu(je->nr_running);
 761         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 762         je->nr_pending          = le32_to_cpu(je->nr_pending);
 763         je->files_open          = le32_to_cpu(je->files_open);
 764         je->m_rate              = le32_to_cpu(je->m_rate);
 765         je->t_rate              = le32_to_cpu(je->t_rate);
 766         je->m_iops              = le32_to_cpu(je->m_iops);
 767         je->t_iops              = le32_to_cpu(je->t_iops);
 768
 769         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 770                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 771                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 772         }
 773
 774         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 775         je->eta_sec             = le64_to_cpu(je->eta_sec);
 776         je->is_pow2             = le32_to_cpu(je->is_pow2);
 777 }
 778
 779 static void sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 780 {
 781         int i;
 782
 783         dst->nr_running         += je->nr_running;
 784         dst->nr_ramp            += je->nr_ramp;
 785         dst->nr_pending         += je->nr_pending;
 786         dst->files_open         += je->files_open;
 787         dst->m_rate             += je->m_rate;
 788         dst->t_rate             += je->t_rate;
 789         dst->m_iops             += je->m_iops;
 790         dst->t_iops             += je->t_iops;
 791
 792         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 793                 dst->rate[i]    += je->rate[i];
 794                 dst->iops[i]    += je->iops[i];
 795         }
 796
 797         dst->elapsed_sec        += je->elapsed_sec;
 798
 799         if (je->eta_sec > dst->eta_sec)
 800                 dst->eta_sec = je->eta_sec;
 801 }
 802
 803 static void dec_jobs_eta(struct client_eta *eta)
 804 {
 805         if (!--eta->pending) {
 806                 display_thread_status(&eta->eta);
 807                 free(eta);
 808         }
 809 }
 810
 811 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 812 {
 813         struct fio_net_int_cmd *icmd = NULL;
 814         struct flist_head *entry;
 815
 816         flist_for_each(entry, &client->cmd_list) {
 817                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
 818
 819                 if (cmd->tag == (uintptr_t) icmd)
 820                         break;
 821
 822                 icmd = NULL;
 823         }
 824
 825         if (!icmd) {
 826                 log_err("fio: client: unable to find matching tag\n");
 827                 return;
 828         }
 829
 830         flist_del(&icmd->list);
 831         cmd->tag = icmd->saved_tag;
 832         free(icmd);
 833 }
 834
 835 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 836 {
 837         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 838         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 839
 840         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 841
 842         assert(client->eta_in_flight == eta);
 843
 844         client->eta_in_flight = NULL;
 845         flist_del_init(&client->eta_list);
 846
 847         convert_jobs_eta(je);
 848         sum_jobs_eta(&eta->eta, je);
 849         dec_jobs_eta(eta);
 850 }
 851
 852 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 853 {
 854         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 855         const char *os, *arch;
 856         char bit[16];
 857
 858         os = fio_get_os_string(probe->os);
 859         if (!os)
 860                 os = "unknown";
 861
 862         arch = fio_get_arch_string(probe->arch);
 863         if (!arch)
 864                 os = "unknown";
 865
 866         sprintf(bit, "%d-bit", probe->bpp * 8);
 867
 868         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s\n",
 869                 probe->hostname, probe->bigendian, bit, os, arch,
 870                 probe->fio_version);
 871
 872         if (!client->name)
 873                 client->name = strdup((char *) probe->hostname);
 874 }
 875
 876 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
 877 {
 878         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 879
 880         client->state = Client_started;
 881         client->jobs = le32_to_cpu(pdu->jobs);
 882         client->nr_stat = le32_to_cpu(pdu->stat_outputs);
 883
 884         if (sum_stat_clients > 1)
 885                 do_output_all_clients = 1;
 886
 887         sum_stat_clients += client->nr_stat;
 888 }
 889
 890 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
 891 {
 892         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
 893
 894         client->state = Client_stopped;
 895         client->error = le32_to_cpu(pdu->error);
 896
 897         if (client->error)
 898                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
 899 }
 900
 901 static int handle_client(struct fio_client *client)
 902 {
 903         struct fio_net_cmd *cmd;
 904
 905         dprint(FD_NET, "client: handle %s\n", client->hostname);
 906
 907         cmd = fio_net_recv_cmd(client->fd);
 908         if (!cmd)
 909                 return 0;
 910
 911         dprint(FD_NET, "client: got cmd op %s from %s\n",
 912                                 fio_server_op(cmd->opcode), client->hostname);
 913
 914         switch (cmd->opcode) {
 915         case FIO_NET_CMD_QUIT:
 916                 remove_client(client);
 917                 free(cmd);
 918                 break;
 919         case FIO_NET_CMD_TEXT: {
 920                 const char *buf = (const char *) cmd->payload;
 921                 const char *name;
 922                 int fio_unused ret;
 923
 924                 name = client->name ? client->name : client->hostname;
 925
 926                 if (!client->skip_newline)
 927                         fprintf(f_out, "<%s> ", name);
 928                 ret = fwrite(buf, cmd->pdu_len, 1, f_out);
 929                 fflush(f_out);
 930                 client->skip_newline = strchr(buf, '\n') == NULL;
 931                 free(cmd);
 932                 break;
 933                 }
 934         case FIO_NET_CMD_DU:
 935                 handle_du(client, cmd);
 936                 free(cmd);
 937                 break;
 938         case FIO_NET_CMD_TS:
 939                 handle_ts(client, cmd);
 940                 free(cmd);
 941                 break;
 942         case FIO_NET_CMD_GS:
 943                 handle_gs(cmd);
 944                 free(cmd);
 945                 break;
 946         case FIO_NET_CMD_ETA:
 947                 remove_reply_cmd(client, cmd);
 948                 handle_eta(client, cmd);
 949                 free(cmd);
 950                 break;
 951         case FIO_NET_CMD_PROBE:
 952                 remove_reply_cmd(client, cmd);
 953                 handle_probe(client, cmd);
 954                 free(cmd);
 955                 break;
 956         case FIO_NET_CMD_RUN:
 957                 client->state = Client_running;
 958                 free(cmd);
 959                 break;
 960         case FIO_NET_CMD_START:
 961                 handle_start(client, cmd);
 962                 free(cmd);
 963                 break;
 964         case FIO_NET_CMD_STOP:
 965                 handle_stop(client, cmd);
 966                 free(cmd);
 967                 break;
 968         default:
 969                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
 970                 free(cmd);
 971                 break;
 972         }
 973
 974         return 1;
 975 }
 976
 977 static void request_client_etas(void)
 978 {
 979         struct fio_client *client;
 980         struct flist_head *entry;
 981         struct client_eta *eta;
 982         int skipped = 0;
 983
 984         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
 985
 986         eta = malloc(sizeof(*eta));
 987         memset(&eta->eta, 0, sizeof(eta->eta));
 988         eta->pending = nr_clients;
 989
 990         flist_for_each(entry, &client_list) {
 991                 client = flist_entry(entry, struct fio_client, list);
 992
 993                 if (!flist_empty(&client->eta_list)) {
 994                         skipped++;
 995                         continue;
 996                 }
 997                 if (client->state != Client_running)
 998                         continue;
 999
1000                 assert(!client->eta_in_flight);
1001                 flist_add_tail(&client->eta_list, &eta_list);
1002                 client->eta_in_flight = eta;
1003                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1004                                         (uintptr_t) eta, &client->cmd_list);
1005         }
1006
1007         while (skipped--)
1008                 dec_jobs_eta(eta);
1009
1010         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1011 }
1012
1013 static int client_check_cmd_timeout(struct fio_client *client,
1014                                     struct timeval *now)
1015 {
1016         struct fio_net_int_cmd *cmd;
1017         struct flist_head *entry, *tmp;
1018         int ret = 0;
1019
1020         flist_for_each_safe(entry, tmp, &client->cmd_list) {
1021                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
1022
1023                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
1024                         continue;
1025
1026                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1027                                                 fio_server_op(cmd->cmd.opcode));
1028                 flist_del(&cmd->list);
1029                 free(cmd);
1030                 ret = 1;
1031         }
1032
1033         return flist_empty(&client->cmd_list) && ret;
1034 }
1035
1036 static int fio_client_timed_out(void)
1037 {
1038         struct fio_client *client;
1039         struct flist_head *entry, *tmp;
1040         struct timeval tv;
1041         int ret = 0;
1042
1043         fio_gettime(&tv, NULL);
1044
1045         flist_for_each_safe(entry, tmp, &client_list) {
1046                 client = flist_entry(entry, struct fio_client, list);
1047
1048                 if (flist_empty(&client->cmd_list))
1049                         continue;
1050
1051                 if (!client_check_cmd_timeout(client, &tv))
1052                         continue;
1053
1054                 log_err("fio: client %s timed out\n", client->hostname);
1055                 remove_client(client);
1056                 ret = 1;
1057         }
1058
1059         return ret;
1060 }
1061
1062 int fio_handle_clients(void)
1063 {
1064         struct pollfd *pfds;
1065         int i, ret = 0, retval = 0;
1066
1067         fio_gettime(&eta_tv, NULL);
1068
1069         pfds = malloc(nr_clients * sizeof(struct pollfd));
1070
1071         init_thread_stat(&client_ts);
1072         init_group_run_stat(&client_gs);
1073
1074         while (!exit_backend && nr_clients) {
1075                 struct flist_head *entry, *tmp;
1076                 struct fio_client *client;
1077
1078                 i = 0;
1079                 flist_for_each_safe(entry, tmp, &client_list) {
1080                         client = flist_entry(entry, struct fio_client, list);
1081
1082                         if (!client->sent_job &&
1083                             flist_empty(&client->cmd_list)) {
1084                                 remove_client(client);
1085                                 continue;
1086                         }
1087
1088                         pfds[i].fd = client->fd;
1089                         pfds[i].events = POLLIN;
1090                         i++;
1091                 }
1092
1093                 if (!nr_clients)
1094                         break;
1095
1096                 assert(i == nr_clients);
1097
1098                 do {
1099                         struct timeval tv;
1100
1101                         fio_gettime(&tv, NULL);
1102                         if (mtime_since(&eta_tv, &tv) >= 900) {
1103                                 request_client_etas();
1104                                 memcpy(&eta_tv, &tv, sizeof(tv));
1105
1106                                 if (fio_client_timed_out())
1107                                         break;
1108                         }
1109
1110                         ret = poll(pfds, nr_clients, 100);
1111                         if (ret < 0) {
1112                                 if (errno == EINTR)
1113                                         continue;
1114                                 log_err("fio: poll clients: %s\n", strerror(errno));
1115                                 break;
1116                         } else if (!ret)
1117                                 continue;
1118                 } while (ret <= 0);
1119
1120                 for (i = 0; i < nr_clients; i++) {
1121                         if (!(pfds[i].revents & POLLIN))
1122                                 continue;
1123
1124                         client = find_client_by_fd(pfds[i].fd);
1125                         if (!client) {
1126                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1127                                 continue;
1128                         }
1129                         if (!handle_client(client)) {
1130                                 log_info("client: host=%s disconnected\n",
1131                                                 client->hostname);
1132                                 remove_client(client);
1133                                 retval = 1;
1134                         } else if (client->error)
1135                                 retval = 1;
1136                         put_client(client);
1137                 }
1138         }
1139
1140         free(pfds);
1141         return retval;
1142 }