client.c

   1 #include <stdio.h>
   2 #include <stdlib.h>
   3 #include <unistd.h>
   4 #include <limits.h>
   5 #include <errno.h>
   6 #include <fcntl.h>
   7 #include <sys/poll.h>
   8 #include <sys/types.h>
   9 #include <sys/stat.h>
  10 #include <sys/wait.h>
  11 #include <sys/socket.h>
  12 #include <sys/un.h>
  13 #include <netinet/in.h>
  14 #include <arpa/inet.h>
  15 #include <netdb.h>
  16 #include <signal.h>
  17
  18 #include "fio.h"
  19 #include "client.h"
  20 #include "server.h"
  21 #include "flist.h"
  22 #include "hash.h"
  23
  24 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
  25 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd);
  26 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd);
  27 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
  28 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd);
  29 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd);
  30
  31 struct client_ops fio_client_ops = {
  32         .text_op        = handle_text,
  33         .disk_util      = handle_du,
  34         .thread_status  = handle_ts,
  35         .group_stats    = handle_gs,
  36         .stop           = handle_stop,
  37         .eta            = display_thread_status,
  38         .probe          = handle_probe,
  39 };
  40
  41 static struct timeval eta_tv;
  42
  43 static FLIST_HEAD(client_list);
  44 static FLIST_HEAD(eta_list);
  45
  46 static FLIST_HEAD(arg_list);
  47
  48 struct thread_stat client_ts;
  49 struct group_run_stats client_gs;
  50 int sum_stat_clients;
  51
  52 static int sum_stat_nr;
  53
  54 #define FIO_CLIENT_HASH_BITS    7
  55 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  56 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  57 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  58
  59 static void fio_client_add_hash(struct fio_client *client)
  60 {
  61         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  62
  63         bucket &= FIO_CLIENT_HASH_MASK;
  64         flist_add(&client->hash_list, &client_hash[bucket]);
  65 }
  66
  67 static void fio_client_remove_hash(struct fio_client *client)
  68 {
  69         if (!flist_empty(&client->hash_list))
  70                 flist_del_init(&client->hash_list);
  71 }
  72
  73 static void fio_init fio_client_hash_init(void)
  74 {
  75         int i;
  76
  77         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
  78                 INIT_FLIST_HEAD(&client_hash[i]);
  79 }
  80
  81 static struct fio_client *find_client_by_fd(int fd)
  82 {
  83         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
  84         struct fio_client *client;
  85         struct flist_head *entry;
  86
  87         flist_for_each(entry, &client_hash[bucket]) {
  88                 client = flist_entry(entry, struct fio_client, hash_list);
  89
  90                 if (client->fd == fd) {
  91                         client->refs++;
  92                         return client;
  93                 }
  94         }
  95
  96         return NULL;
  97 }
  98
  99 static void remove_client(struct fio_client *client)
 100 {
 101         assert(client->refs);
 102
 103         if (--client->refs)
 104                 return;
 105
 106         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 107         flist_del(&client->list);
 108
 109         fio_client_remove_hash(client);
 110
 111         if (!flist_empty(&client->eta_list)) {
 112                 flist_del_init(&client->eta_list);
 113                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
 114         }
 115
 116         free(client->hostname);
 117         if (client->argv)
 118                 free(client->argv);
 119         if (client->name)
 120                 free(client->name);
 121
 122         free(client);
 123         nr_clients--;
 124         sum_stat_clients--;
 125 }
 126
 127 static void put_client(struct fio_client *client)
 128 {
 129         remove_client(client);
 130 }
 131
 132 static void __fio_client_add_cmd_option(struct fio_client *client,
 133                                         const char *opt)
 134 {
 135         int index;
 136
 137         index = client->argc++;
 138         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 139         client->argv[index] = strdup(opt);
 140         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 141 }
 142
 143 void fio_client_add_cmd_option(void *cookie, const char *opt)
 144 {
 145         struct fio_client *client = cookie;
 146         struct flist_head *entry;
 147
 148         if (!client || !opt)
 149                 return;
 150
 151         __fio_client_add_cmd_option(client, opt);
 152
 153         /*
 154          * Duplicate arguments to shared client group
 155          */
 156         flist_for_each(entry, &arg_list) {
 157                 client = flist_entry(entry, struct fio_client, arg_list);
 158
 159                 __fio_client_add_cmd_option(client, opt);
 160         }
 161 }
 162
 163 struct fio_client *fio_client_add_explicit(struct client_ops *ops,
 164                                            const char *hostname, int type,
 165                                            int port)
 166 {
 167         struct fio_client *client;
 168
 169         client = malloc(sizeof(*client));
 170         memset(client, 0, sizeof(*client));
 171
 172         INIT_FLIST_HEAD(&client->list);
 173         INIT_FLIST_HEAD(&client->hash_list);
 174         INIT_FLIST_HEAD(&client->arg_list);
 175         INIT_FLIST_HEAD(&client->eta_list);
 176         INIT_FLIST_HEAD(&client->cmd_list);
 177
 178         client->hostname = strdup(hostname);
 179
 180         if (type == Fio_client_socket)
 181                 client->is_sock = 1;
 182         else {
 183                 int ipv6;
 184
 185                 ipv6 = type == Fio_client_ipv6;
 186                 if (fio_server_parse_host(hostname, &ipv6,
 187                                                 &client->addr.sin_addr,
 188                                                 &client->addr6.sin6_addr))
 189                         goto err;
 190
 191                 client->port = port;
 192         }
 193
 194         client->fd = -1;
 195         client->ops = ops;
 196         client->refs = 1;
 197
 198         __fio_client_add_cmd_option(client, "fio");
 199
 200         flist_add(&client->list, &client_list);
 201         nr_clients++;
 202         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 203         return client;
 204 err:
 205         free(client);
 206         return NULL;
 207 }
 208
 209 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
 210 {
 211         struct fio_client *existing = *cookie;
 212         struct fio_client *client;
 213
 214         if (existing) {
 215                 /*
 216                  * We always add our "exec" name as the option, hence 1
 217                  * means empty.
 218                  */
 219                 if (existing->argc == 1)
 220                         flist_add_tail(&existing->arg_list, &arg_list);
 221                 else {
 222                         while (!flist_empty(&arg_list))
 223                                 flist_del_init(arg_list.next);
 224                 }
 225         }
 226
 227         client = malloc(sizeof(*client));
 228         memset(client, 0, sizeof(*client));
 229
 230         INIT_FLIST_HEAD(&client->list);
 231         INIT_FLIST_HEAD(&client->hash_list);
 232         INIT_FLIST_HEAD(&client->arg_list);
 233         INIT_FLIST_HEAD(&client->eta_list);
 234         INIT_FLIST_HEAD(&client->cmd_list);
 235
 236         if (fio_server_parse_string(hostname, &client->hostname,
 237                                         &client->is_sock, &client->port,
 238                                         &client->addr.sin_addr,
 239                                         &client->addr6.sin6_addr,
 240                                         &client->ipv6))
 241                 return -1;
 242
 243         client->fd = -1;
 244         client->ops = ops;
 245         client->refs = 1;
 246
 247         __fio_client_add_cmd_option(client, "fio");
 248
 249         flist_add(&client->list, &client_list);
 250         nr_clients++;
 251         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 252         *cookie = client;
 253         return 0;
 254 }
 255
 256 static int fio_client_connect_ip(struct fio_client *client)
 257 {
 258         struct sockaddr *addr;
 259         fio_socklen_t socklen;
 260         int fd, domain;
 261
 262         if (client->ipv6) {
 263                 client->addr6.sin6_family = AF_INET6;
 264                 client->addr6.sin6_port = htons(client->port);
 265                 domain = AF_INET6;
 266                 addr = (struct sockaddr *) &client->addr6;
 267                 socklen = sizeof(client->addr6);
 268         } else {
 269                 client->addr.sin_family = AF_INET;
 270                 client->addr.sin_port = htons(client->port);
 271                 domain = AF_INET;
 272                 addr = (struct sockaddr *) &client->addr;
 273                 socklen = sizeof(client->addr);
 274         }
 275
 276         fd = socket(domain, SOCK_STREAM, 0);
 277         if (fd < 0) {
 278                 int ret = -errno;
 279
 280                 log_err("fio: socket: %s\n", strerror(errno));
 281                 return ret;
 282         }
 283
 284         if (connect(fd, addr, socklen) < 0) {
 285                 int ret = -errno;
 286
 287                 log_err("fio: connect: %s\n", strerror(errno));
 288                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 289                                                                 client->port);
 290                 close(fd);
 291                 return ret;
 292         }
 293
 294         return fd;
 295 }
 296
 297 static int fio_client_connect_sock(struct fio_client *client)
 298 {
 299         struct sockaddr_un *addr = &client->addr_un;
 300         fio_socklen_t len;
 301         int fd;
 302
 303         memset(addr, 0, sizeof(*addr));
 304         addr->sun_family = AF_UNIX;
 305         strcpy(addr->sun_path, client->hostname);
 306
 307         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 308         if (fd < 0) {
 309                 int ret = -errno;
 310
 311                 log_err("fio: socket: %s\n", strerror(errno));
 312                 return ret;
 313         }
 314
 315         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 316         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 317                 int ret = -errno;
 318
 319                 log_err("fio: connect; %s\n", strerror(errno));
 320                 close(fd);
 321                 return ret;
 322         }
 323
 324         return fd;
 325 }
 326
 327 int fio_client_connect(struct fio_client *client)
 328 {
 329         int fd;
 330
 331         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 332
 333         if (client->is_sock)
 334                 fd = fio_client_connect_sock(client);
 335         else
 336                 fd = fio_client_connect_ip(client);
 337
 338         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 339
 340         if (fd < 0)
 341                 return fd;
 342
 343         client->fd = fd;
 344         fio_client_add_hash(client);
 345         client->state = Client_connected;
 346         return 0;
 347 }
 348
 349 void fio_client_terminate(struct fio_client *client)
 350 {
 351         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0, NULL);
 352 }
 353
 354 void fio_clients_terminate(void)
 355 {
 356         struct flist_head *entry;
 357         struct fio_client *client;
 358
 359         dprint(FD_NET, "client: terminate clients\n");
 360
 361         flist_for_each(entry, &client_list) {
 362                 client = flist_entry(entry, struct fio_client, list);
 363                 fio_client_terminate(client);
 364         }
 365 }
 366
 367 static void sig_int(int sig)
 368 {
 369         dprint(FD_NET, "client: got signal %d\n", sig);
 370         fio_clients_terminate();
 371 }
 372
 373 static void client_signal_handler(void)
 374 {
 375         struct sigaction act;
 376
 377         memset(&act, 0, sizeof(act));
 378         act.sa_handler = sig_int;
 379         act.sa_flags = SA_RESTART;
 380         sigaction(SIGINT, &act, NULL);
 381
 382         memset(&act, 0, sizeof(act));
 383         act.sa_handler = sig_int;
 384         act.sa_flags = SA_RESTART;
 385         sigaction(SIGTERM, &act, NULL);
 386 }
 387
 388 static void probe_client(struct fio_client *client)
 389 {
 390         dprint(FD_NET, "client: send probe\n");
 391
 392         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
 393 }
 394
 395 static int send_client_cmd_line(struct fio_client *client)
 396 {
 397         struct cmd_single_line_pdu *cslp;
 398         struct cmd_line_pdu *clp;
 399         unsigned long offset;
 400         unsigned int *lens;
 401         void *pdu;
 402         size_t mem;
 403         int i, ret;
 404
 405         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 406
 407         lens = malloc(client->argc * sizeof(unsigned int));
 408
 409         /*
 410          * Find out how much mem we need
 411          */
 412         for (i = 0, mem = 0; i < client->argc; i++) {
 413                 lens[i] = strlen(client->argv[i]) + 1;
 414                 mem += lens[i];
 415         }
 416
 417         /*
 418          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 419          */
 420         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 421
 422         pdu = malloc(mem);
 423         clp = pdu;
 424         offset = sizeof(*clp);
 425
 426         for (i = 0; i < client->argc; i++) {
 427                 uint16_t arg_len = lens[i];
 428
 429                 cslp = pdu + offset;
 430                 strcpy((char *) cslp->text, client->argv[i]);
 431                 cslp->len = cpu_to_le16(arg_len);
 432                 offset += sizeof(*cslp) + arg_len;
 433         }
 434
 435         free(lens);
 436         clp->lines = cpu_to_le16(client->argc);
 437         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
 438         free(pdu);
 439         return ret;
 440 }
 441
 442 int fio_clients_connect(void)
 443 {
 444         struct fio_client *client;
 445         struct flist_head *entry, *tmp;
 446         int ret;
 447
 448 #ifdef WIN32
 449         WSADATA wsd;
 450         WSAStartup(MAKEWORD(2,2), &wsd);
 451 #endif
 452
 453         dprint(FD_NET, "client: connect all\n");
 454
 455         client_signal_handler();
 456
 457         flist_for_each_safe(entry, tmp, &client_list) {
 458                 client = flist_entry(entry, struct fio_client, list);
 459
 460                 ret = fio_client_connect(client);
 461                 if (ret) {
 462                         remove_client(client);
 463                         continue;
 464                 }
 465
 466                 probe_client(client);
 467
 468                 if (client->argc > 1)
 469                         send_client_cmd_line(client);
 470         }
 471
 472         return !nr_clients;
 473 }
 474
 475 int fio_start_client(struct fio_client *client)
 476 {
 477         dprint(FD_NET, "client: start %s\n", client->hostname);
 478         return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
 479 }
 480
 481 int fio_start_all_clients(void)
 482 {
 483         struct fio_client *client;
 484         struct flist_head *entry, *tmp;
 485         int ret;
 486
 487         dprint(FD_NET, "client: start all\n");
 488
 489         flist_for_each_safe(entry, tmp, &client_list) {
 490                 client = flist_entry(entry, struct fio_client, list);
 491
 492                 ret = fio_start_client(client);
 493                 if (ret) {
 494                         remove_client(client);
 495                         continue;
 496                 }
 497         }
 498
 499         return flist_empty(&client_list);
 500 }
 501
 502 /*
 503  * Send file contents to server backend. We could use sendfile(), but to remain
 504  * more portable lets just read/write the darn thing.
 505  */
 506 static int __fio_client_send_ini(struct fio_client *client, const char *filename)
 507 {
 508         struct stat sb;
 509         char *p, *buf;
 510         off_t len;
 511         int fd, ret;
 512
 513         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 514
 515         fd = open(filename, O_RDONLY);
 516         if (fd < 0) {
 517                 int ret = -errno;
 518
 519                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 520                 return ret;
 521         }
 522
 523         if (fstat(fd, &sb) < 0) {
 524                 int ret = -errno;
 525
 526                 log_err("fio: job file stat: %s\n", strerror(errno));
 527                 close(fd);
 528                 return ret;
 529         }
 530
 531         buf = malloc(sb.st_size);
 532
 533         len = sb.st_size;
 534         p = buf;
 535         do {
 536                 ret = read(fd, p, len);
 537                 if (ret > 0) {
 538                         len -= ret;
 539                         if (!len)
 540                                 break;
 541                         p += ret;
 542                         continue;
 543                 } else if (!ret)
 544                         break;
 545                 else if (errno == EAGAIN || errno == EINTR)
 546                         continue;
 547         } while (1);
 548
 549         if (len) {
 550                 log_err("fio: failed reading job file %s\n", filename);
 551                 close(fd);
 552                 free(buf);
 553                 return 1;
 554         }
 555
 556         client->sent_job = 1;
 557         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size, 0);
 558         free(buf);
 559         close(fd);
 560         return ret;
 561 }
 562
 563 int fio_client_send_ini(struct fio_client *client, const char *filename)
 564 {
 565         int ret;
 566
 567         ret = __fio_client_send_ini(client, filename);
 568         if (ret) {
 569                 remove_client(client);
 570                 return ret;
 571         }
 572
 573         client->sent_job = 1;
 574         return ret;
 575 }
 576
 577 int fio_clients_send_ini(const char *filename)
 578 {
 579         struct fio_client *client;
 580         struct flist_head *entry, *tmp;
 581
 582         flist_for_each_safe(entry, tmp, &client_list) {
 583                 client = flist_entry(entry, struct fio_client, list);
 584
 585                 fio_client_send_ini(client, filename);
 586         }
 587
 588         return !nr_clients;
 589 }
 590
 591 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 592 {
 593         dst->max_val    = le64_to_cpu(src->max_val);
 594         dst->min_val    = le64_to_cpu(src->min_val);
 595         dst->samples    = le64_to_cpu(src->samples);
 596
 597         /*
 598          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 599          */
 600         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 601         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 602 }
 603
 604 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 605 {
 606         int i, j;
 607
 608         dst->error      = le32_to_cpu(src->error);
 609         dst->groupid    = le32_to_cpu(src->groupid);
 610         dst->pid        = le32_to_cpu(src->pid);
 611         dst->members    = le32_to_cpu(src->members);
 612
 613         for (i = 0; i < 2; i++) {
 614                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 615                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 616                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 617                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 618         }
 619
 620         dst->usr_time           = le64_to_cpu(src->usr_time);
 621         dst->sys_time           = le64_to_cpu(src->sys_time);
 622         dst->ctx                = le64_to_cpu(src->ctx);
 623         dst->minf               = le64_to_cpu(src->minf);
 624         dst->majf               = le64_to_cpu(src->majf);
 625         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 626
 627         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 628                 fio_fp64_t *fps = &src->percentile_list[i];
 629                 fio_fp64_t *fpd = &dst->percentile_list[i];
 630
 631                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 632         }
 633
 634         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 635                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 636                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 637                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 638         }
 639
 640         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 641                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 642                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 643         }
 644
 645         for (i = 0; i < 2; i++)
 646                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 647                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 648
 649         for (i = 0; i < 3; i++) {
 650                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 651                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 652         }
 653
 654         dst->total_submit       = le64_to_cpu(src->total_submit);
 655         dst->total_complete     = le64_to_cpu(src->total_complete);
 656
 657         for (i = 0; i < 2; i++) {
 658                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 659                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 660         }
 661
 662         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 663         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 664         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 665         dst->first_error        = le32_to_cpu(src->first_error);
 666         dst->kb_base            = le32_to_cpu(src->kb_base);
 667 }
 668
 669 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 670 {
 671         int i;
 672
 673         for (i = 0; i < 2; i++) {
 674                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 675                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 676                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 677                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 678                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 679                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 680         }
 681
 682         dst->kb_base    = le32_to_cpu(src->kb_base);
 683         dst->groupid    = le32_to_cpu(src->groupid);
 684 }
 685
 686 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
 687 {
 688         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 689
 690         show_thread_status(&p->ts, &p->rs);
 691
 692         if (sum_stat_clients == 1)
 693                 return;
 694
 695         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 696         sum_group_stats(&client_gs, &p->rs);
 697
 698         client_ts.members++;
 699         client_ts.groupid = p->ts.groupid;
 700
 701         if (++sum_stat_nr == sum_stat_clients) {
 702                 strcpy(client_ts.name, "All clients");
 703                 show_thread_status(&client_ts, &client_gs);
 704         }
 705 }
 706
 707 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
 708 {
 709         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 710
 711         show_group_stats(gs);
 712 }
 713
 714 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
 715 {
 716         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
 717         const char *buf = (const char *) pdu->buf;
 718         const char *name;
 719         int fio_unused ret;
 720
 721         name = client->name ? client->name : client->hostname;
 722
 723         if (!client->skip_newline)
 724                 fprintf(f_out, "<%s> ", name);
 725         ret = fwrite(buf, pdu->buf_len, 1, f_out);
 726         fflush(f_out);
 727         client->skip_newline = strchr(buf, '\n') == NULL;
 728 }
 729
 730 static void convert_agg(struct disk_util_agg *agg)
 731 {
 732         int i;
 733
 734         for (i = 0; i < 2; i++) {
 735                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 736                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 737                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 738                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 739         }
 740
 741         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 742         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 743         agg->slavecount         = le32_to_cpu(agg->slavecount);
 744         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 745 }
 746
 747 static void convert_dus(struct disk_util_stat *dus)
 748 {
 749         int i;
 750
 751         for (i = 0; i < 2; i++) {
 752                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
 753                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
 754                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
 755                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
 756         }
 757
 758         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
 759         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
 760         dus->msec               = le64_to_cpu(dus->msec);
 761 }
 762
 763 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 764 {
 765         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 766
 767         if (!client->disk_stats_shown) {
 768                 client->disk_stats_shown = 1;
 769                 log_info("\nDisk stats (read/write):\n");
 770         }
 771
 772         print_disk_util(&du->dus, &du->agg, terse_output);
 773 }
 774
 775 static void convert_jobs_eta(struct jobs_eta *je)
 776 {
 777         int i;
 778
 779         je->nr_running          = le32_to_cpu(je->nr_running);
 780         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 781         je->nr_pending          = le32_to_cpu(je->nr_pending);
 782         je->files_open          = le32_to_cpu(je->files_open);
 783
 784         for (i = 0; i < 2; i++) {
 785                 je->m_rate[i]           = le32_to_cpu(je->m_rate[i]);
 786                 je->t_rate[i]           = le32_to_cpu(je->t_rate[i]);
 787                 je->m_iops[i]           = le32_to_cpu(je->m_iops[i]);
 788                 je->t_iops[i]           = le32_to_cpu(je->t_iops[i]);
 789                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 790                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 791         }
 792
 793         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 794         je->eta_sec             = le64_to_cpu(je->eta_sec);
 795         je->nr_threads          = le32_to_cpu(je->nr_threads);
 796 }
 797
 798 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 799 {
 800         int i;
 801
 802         dst->nr_running         += je->nr_running;
 803         dst->nr_ramp            += je->nr_ramp;
 804         dst->nr_pending         += je->nr_pending;
 805         dst->files_open         += je->files_open;
 806
 807         for (i = 0; i < 2; i++) {
 808                 dst->m_rate[i]  += je->m_rate[i];
 809                 dst->t_rate[i]  += je->t_rate[i];
 810                 dst->m_iops[i]  += je->m_iops[i];
 811                 dst->t_iops[i]  += je->t_iops[i];
 812                 dst->rate[i]    += je->rate[i];
 813                 dst->iops[i]    += je->iops[i];
 814         }
 815
 816         dst->elapsed_sec        += je->elapsed_sec;
 817
 818         if (je->eta_sec > dst->eta_sec)
 819                 dst->eta_sec = je->eta_sec;
 820
 821         dst->nr_threads         += je->nr_threads;
 822         /* we need to handle je->run_str too ... */
 823 }
 824
 825 void fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
 826 {
 827         if (!--eta->pending) {
 828                 eta_fn(&eta->eta);
 829                 free(eta);
 830         }
 831 }
 832
 833 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 834 {
 835         struct fio_net_int_cmd *icmd = NULL;
 836         struct flist_head *entry;
 837
 838         flist_for_each(entry, &client->cmd_list) {
 839                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
 840
 841                 if (cmd->tag == (uintptr_t) icmd)
 842                         break;
 843
 844                 icmd = NULL;
 845         }
 846
 847         if (!icmd) {
 848                 log_err("fio: client: unable to find matching tag\n");
 849                 return;
 850         }
 851
 852         flist_del(&icmd->list);
 853         cmd->tag = icmd->saved_tag;
 854         free(icmd);
 855 }
 856
 857 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 858 {
 859         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 860         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 861
 862         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 863
 864         assert(client->eta_in_flight == eta);
 865
 866         client->eta_in_flight = NULL;
 867         flist_del_init(&client->eta_list);
 868
 869         if (client->ops->jobs_eta)
 870                 client->ops->jobs_eta(client, je);
 871
 872         fio_client_sum_jobs_eta(&eta->eta, je);
 873         fio_client_dec_jobs_eta(eta, client->ops->eta);
 874 }
 875
 876 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 877 {
 878         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 879         const char *os, *arch;
 880         char bit[16];
 881
 882         os = fio_get_os_string(probe->os);
 883         if (!os)
 884                 os = "unknown";
 885
 886         arch = fio_get_arch_string(probe->arch);
 887         if (!arch)
 888                 os = "unknown";
 889
 890         sprintf(bit, "%d-bit", probe->bpp * 8);
 891
 892         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
 893                 probe->hostname, probe->bigendian, bit, os, arch,
 894                 probe->fio_major, probe->fio_minor, probe->fio_patch);
 895
 896         if (!client->name)
 897                 client->name = strdup((char *) probe->hostname);
 898 }
 899
 900 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
 901 {
 902         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 903
 904         client->state = Client_started;
 905         client->jobs = le32_to_cpu(pdu->jobs);
 906 }
 907
 908 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
 909 {
 910         if (client->error)
 911                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
 912 }
 913
 914 static void convert_stop(struct fio_net_cmd *cmd)
 915 {
 916         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
 917
 918         pdu->error = le32_to_cpu(pdu->error);
 919 }
 920
 921 static void convert_text(struct fio_net_cmd *cmd)
 922 {
 923         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
 924
 925         pdu->level      = le32_to_cpu(pdu->level);
 926         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
 927         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
 928         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
 929 }
 930
 931 int fio_handle_client(struct fio_client *client)
 932 {
 933         struct client_ops *ops = client->ops;
 934         struct fio_net_cmd *cmd;
 935
 936         dprint(FD_NET, "client: handle %s\n", client->hostname);
 937
 938         cmd = fio_net_recv_cmd(client->fd);
 939         if (!cmd)
 940                 return 0;
 941
 942         dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
 943                 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
 944
 945         switch (cmd->opcode) {
 946         case FIO_NET_CMD_QUIT:
 947                 if (ops->quit)
 948                         ops->quit(client);
 949                 remove_client(client);
 950                 free(cmd);
 951                 break;
 952         case FIO_NET_CMD_TEXT:
 953                 convert_text(cmd);
 954                 ops->text_op(client, cmd);
 955                 free(cmd);
 956                 break;
 957         case FIO_NET_CMD_DU: {
 958                 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 959
 960                 convert_dus(&du->dus);
 961                 convert_agg(&du->agg);
 962
 963                 ops->disk_util(client, cmd);
 964                 free(cmd);
 965                 break;
 966                 }
 967         case FIO_NET_CMD_TS: {
 968                 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 969
 970                 convert_ts(&p->ts, &p->ts);
 971                 convert_gs(&p->rs, &p->rs);
 972
 973                 ops->thread_status(client, cmd);
 974                 free(cmd);
 975                 break;
 976                 }
 977         case FIO_NET_CMD_GS: {
 978                 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 979
 980                 convert_gs(gs, gs);
 981
 982                 ops->group_stats(client, cmd);
 983                 free(cmd);
 984                 break;
 985                 }
 986         case FIO_NET_CMD_ETA: {
 987                 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 988
 989                 remove_reply_cmd(client, cmd);
 990                 convert_jobs_eta(je);
 991                 handle_eta(client, cmd);
 992                 free(cmd);
 993                 break;
 994                 }
 995         case FIO_NET_CMD_PROBE:
 996                 remove_reply_cmd(client, cmd);
 997                 ops->probe(client, cmd);
 998                 free(cmd);
 999                 break;
1000         case FIO_NET_CMD_SERVER_START:
1001                 client->state = Client_running;
1002                 free(cmd);
1003                 break;
1004         case FIO_NET_CMD_START:
1005                 handle_start(client, cmd);
1006                 free(cmd);
1007                 break;
1008         case FIO_NET_CMD_STOP: {
1009                 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1010
1011                 convert_stop(cmd);
1012                 client->state = Client_stopped;
1013                 client->error = pdu->error;
1014                 ops->stop(client, cmd);
1015                 free(cmd);
1016                 break;
1017                 }
1018         case FIO_NET_CMD_ADD_JOB:
1019                 if (ops->add_job)
1020                         ops->add_job(client, cmd);
1021                 free(cmd);
1022                 break;
1023         default:
1024                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1025                 free(cmd);
1026                 break;
1027         }
1028
1029         return 1;
1030 }
1031
1032 static void request_client_etas(struct client_ops *ops)
1033 {
1034         struct fio_client *client;
1035         struct flist_head *entry;
1036         struct client_eta *eta;
1037         int skipped = 0;
1038
1039         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1040
1041         eta = malloc(sizeof(*eta));
1042         memset(&eta->eta, 0, sizeof(eta->eta));
1043         eta->pending = nr_clients;
1044
1045         flist_for_each(entry, &client_list) {
1046                 client = flist_entry(entry, struct fio_client, list);
1047
1048                 if (!flist_empty(&client->eta_list)) {
1049                         skipped++;
1050                         continue;
1051                 }
1052                 if (client->state != Client_running)
1053                         continue;
1054
1055                 assert(!client->eta_in_flight);
1056                 flist_add_tail(&client->eta_list, &eta_list);
1057                 client->eta_in_flight = eta;
1058                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1059                                         (uintptr_t) eta, &client->cmd_list);
1060         }
1061
1062         while (skipped--)
1063                 fio_client_dec_jobs_eta(eta, ops->eta);
1064
1065         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1066 }
1067
1068 static int client_check_cmd_timeout(struct fio_client *client,
1069                                     struct timeval *now)
1070 {
1071         struct fio_net_int_cmd *cmd;
1072         struct flist_head *entry, *tmp;
1073         int ret = 0;
1074
1075         flist_for_each_safe(entry, tmp, &client->cmd_list) {
1076                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
1077
1078                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
1079                         continue;
1080
1081                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1082                                                 fio_server_op(cmd->cmd.opcode));
1083                 flist_del(&cmd->list);
1084                 free(cmd);
1085                 ret = 1;
1086         }
1087
1088         return flist_empty(&client->cmd_list) && ret;
1089 }
1090
1091 static int fio_check_clients_timed_out(void)
1092 {
1093         struct fio_client *client;
1094         struct flist_head *entry, *tmp;
1095         struct timeval tv;
1096         int ret = 0;
1097
1098         gettimeofday(&tv, NULL);
1099
1100         flist_for_each_safe(entry, tmp, &client_list) {
1101                 client = flist_entry(entry, struct fio_client, list);
1102
1103                 if (flist_empty(&client->cmd_list))
1104                         continue;
1105
1106                 if (!client_check_cmd_timeout(client, &tv))
1107                         continue;
1108
1109                 if (client->ops->timed_out)
1110                         client->ops->timed_out(client);
1111                 else
1112                         log_err("fio: client %s timed out\n", client->hostname);
1113
1114                 remove_client(client);
1115                 ret = 1;
1116         }
1117
1118         return ret;
1119 }
1120
1121 int fio_handle_clients(struct client_ops *ops)
1122 {
1123         struct pollfd *pfds;
1124         int i, ret = 0, retval = 0;
1125
1126         gettimeofday(&eta_tv, NULL);
1127
1128         pfds = malloc(nr_clients * sizeof(struct pollfd));
1129
1130         sum_stat_clients = nr_clients;
1131         init_thread_stat(&client_ts);
1132         init_group_run_stat(&client_gs);
1133
1134         while (!exit_backend && nr_clients) {
1135                 struct flist_head *entry, *tmp;
1136                 struct fio_client *client;
1137
1138                 i = 0;
1139                 flist_for_each_safe(entry, tmp, &client_list) {
1140                         client = flist_entry(entry, struct fio_client, list);
1141
1142                         if (!client->sent_job && !client->ops->stay_connected &&
1143                             flist_empty(&client->cmd_list)) {
1144                                 remove_client(client);
1145                                 continue;
1146                         }
1147
1148                         pfds[i].fd = client->fd;
1149                         pfds[i].events = POLLIN;
1150                         i++;
1151                 }
1152
1153                 if (!nr_clients)
1154                         break;
1155
1156                 assert(i == nr_clients);
1157
1158                 do {
1159                         struct timeval tv;
1160
1161                         gettimeofday(&tv, NULL);
1162                         if (mtime_since(&eta_tv, &tv) >= 900) {
1163                                 request_client_etas(ops);
1164                                 memcpy(&eta_tv, &tv, sizeof(tv));
1165
1166                                 if (fio_check_clients_timed_out())
1167                                         break;
1168                         }
1169
1170                         ret = poll(pfds, nr_clients, 100);
1171                         if (ret < 0) {
1172                                 if (errno == EINTR)
1173                                         continue;
1174                                 log_err("fio: poll clients: %s\n", strerror(errno));
1175                                 break;
1176                         } else if (!ret)
1177                                 continue;
1178                 } while (ret <= 0);
1179
1180                 for (i = 0; i < nr_clients; i++) {
1181                         if (!(pfds[i].revents & POLLIN))
1182                                 continue;
1183
1184                         client = find_client_by_fd(pfds[i].fd);
1185                         if (!client) {
1186                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1187                                 continue;
1188                         }
1189                         if (!fio_handle_client(client)) {
1190                                 log_info("client: host=%s disconnected\n",
1191                                                 client->hostname);
1192                                 remove_client(client);
1193                                 retval = 1;
1194                         } else if (client->error)
1195                                 retval = 1;
1196                         put_client(client);
1197                 }
1198         }
1199
1200         free(pfds);
1201         return retval;
1202 }