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 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd);
  31
  32 struct client_ops fio_client_ops = {
  33         .text           = handle_text,
  34         .disk_util      = handle_du,
  35         .thread_status  = handle_ts,
  36         .group_stats    = handle_gs,
  37         .stop           = handle_stop,
  38         .start          = handle_start,
  39         .eta            = display_thread_status,
  40         .probe          = handle_probe,
  41         .eta_msec       = FIO_CLIENT_DEF_ETA_MSEC,
  42         .client_type    = FIO_CLIENT_TYPE_CLI,
  43 };
  44
  45 static struct timeval eta_tv;
  46
  47 static FLIST_HEAD(client_list);
  48 static FLIST_HEAD(eta_list);
  49
  50 static FLIST_HEAD(arg_list);
  51
  52 struct thread_stat client_ts;
  53 struct group_run_stats client_gs;
  54 int sum_stat_clients;
  55
  56 static int sum_stat_nr;
  57
  58 #define FIO_CLIENT_HASH_BITS    7
  59 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  60 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  61 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  62
  63 static void fio_client_add_hash(struct fio_client *client)
  64 {
  65         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  66
  67         bucket &= FIO_CLIENT_HASH_MASK;
  68         flist_add(&client->hash_list, &client_hash[bucket]);
  69 }
  70
  71 static void fio_client_remove_hash(struct fio_client *client)
  72 {
  73         if (!flist_empty(&client->hash_list))
  74                 flist_del_init(&client->hash_list);
  75 }
  76
  77 static void fio_init fio_client_hash_init(void)
  78 {
  79         int i;
  80
  81         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
  82                 INIT_FLIST_HEAD(&client_hash[i]);
  83 }
  84
  85 static struct fio_client *find_client_by_fd(int fd)
  86 {
  87         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
  88         struct fio_client *client;
  89         struct flist_head *entry;
  90
  91         flist_for_each(entry, &client_hash[bucket]) {
  92                 client = flist_entry(entry, struct fio_client, hash_list);
  93
  94                 if (client->fd == fd) {
  95                         client->refs++;
  96                         return client;
  97                 }
  98         }
  99
 100         return NULL;
 101 }
 102
 103 void fio_put_client(struct fio_client *client)
 104 {
 105         if (--client->refs)
 106                 return;
 107
 108         free(client->hostname);
 109         if (client->argv)
 110                 free(client->argv);
 111         if (client->name)
 112                 free(client->name);
 113
 114         free(client);
 115 }
 116
 117 static void remove_client(struct fio_client *client)
 118 {
 119         assert(client->refs);
 120
 121         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 122
 123         if (!flist_empty(&client->list))
 124                 flist_del_init(&client->list);
 125
 126         fio_client_remove_hash(client);
 127
 128         if (!flist_empty(&client->eta_list)) {
 129                 flist_del_init(&client->eta_list);
 130                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
 131         }
 132
 133         nr_clients--;
 134         sum_stat_clients--;
 135
 136         fio_put_client(client);
 137 }
 138
 139 struct fio_client *fio_get_client(struct fio_client *client)
 140 {
 141         client->refs++;
 142         return client;
 143 }
 144
 145 static void __fio_client_add_cmd_option(struct fio_client *client,
 146                                         const char *opt)
 147 {
 148         int index;
 149
 150         index = client->argc++;
 151         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 152         client->argv[index] = strdup(opt);
 153         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 154 }
 155
 156 void fio_client_add_cmd_option(void *cookie, const char *opt)
 157 {
 158         struct fio_client *client = cookie;
 159         struct flist_head *entry;
 160
 161         if (!client || !opt)
 162                 return;
 163
 164         __fio_client_add_cmd_option(client, opt);
 165
 166         /*
 167          * Duplicate arguments to shared client group
 168          */
 169         flist_for_each(entry, &arg_list) {
 170                 client = flist_entry(entry, struct fio_client, arg_list);
 171
 172                 __fio_client_add_cmd_option(client, opt);
 173         }
 174 }
 175
 176 struct fio_client *fio_client_add_explicit(struct client_ops *ops,
 177                                            const char *hostname, int type,
 178                                            int port)
 179 {
 180         struct fio_client *client;
 181
 182         client = malloc(sizeof(*client));
 183         memset(client, 0, sizeof(*client));
 184
 185         INIT_FLIST_HEAD(&client->list);
 186         INIT_FLIST_HEAD(&client->hash_list);
 187         INIT_FLIST_HEAD(&client->arg_list);
 188         INIT_FLIST_HEAD(&client->eta_list);
 189         INIT_FLIST_HEAD(&client->cmd_list);
 190
 191         client->hostname = strdup(hostname);
 192
 193         if (type == Fio_client_socket)
 194                 client->is_sock = 1;
 195         else {
 196                 int ipv6;
 197
 198                 ipv6 = type == Fio_client_ipv6;
 199                 if (fio_server_parse_host(hostname, &ipv6,
 200                                                 &client->addr.sin_addr,
 201                                                 &client->addr6.sin6_addr))
 202                         goto err;
 203
 204                 client->port = port;
 205         }
 206
 207         client->fd = -1;
 208         client->ops = ops;
 209         client->refs = 1;
 210         client->type = ops->client_type;
 211
 212         __fio_client_add_cmd_option(client, "fio");
 213
 214         flist_add(&client->list, &client_list);
 215         nr_clients++;
 216         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 217         return client;
 218 err:
 219         free(client);
 220         return NULL;
 221 }
 222
 223 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
 224 {
 225         struct fio_client *existing = *cookie;
 226         struct fio_client *client;
 227
 228         if (existing) {
 229                 /*
 230                  * We always add our "exec" name as the option, hence 1
 231                  * means empty.
 232                  */
 233                 if (existing->argc == 1)
 234                         flist_add_tail(&existing->arg_list, &arg_list);
 235                 else {
 236                         while (!flist_empty(&arg_list))
 237                                 flist_del_init(arg_list.next);
 238                 }
 239         }
 240
 241         client = malloc(sizeof(*client));
 242         memset(client, 0, sizeof(*client));
 243
 244         INIT_FLIST_HEAD(&client->list);
 245         INIT_FLIST_HEAD(&client->hash_list);
 246         INIT_FLIST_HEAD(&client->arg_list);
 247         INIT_FLIST_HEAD(&client->eta_list);
 248         INIT_FLIST_HEAD(&client->cmd_list);
 249
 250         if (fio_server_parse_string(hostname, &client->hostname,
 251                                         &client->is_sock, &client->port,
 252                                         &client->addr.sin_addr,
 253                                         &client->addr6.sin6_addr,
 254                                         &client->ipv6))
 255                 return -1;
 256
 257         client->fd = -1;
 258         client->ops = ops;
 259         client->refs = 1;
 260         client->type = ops->client_type;
 261
 262         __fio_client_add_cmd_option(client, "fio");
 263
 264         flist_add(&client->list, &client_list);
 265         nr_clients++;
 266         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 267         *cookie = client;
 268         return 0;
 269 }
 270
 271 static void probe_client(struct fio_client *client)
 272 {
 273         dprint(FD_NET, "client: send probe\n");
 274
 275         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
 276 }
 277
 278 static int fio_client_connect_ip(struct fio_client *client)
 279 {
 280         struct sockaddr *addr;
 281         fio_socklen_t socklen;
 282         int fd, domain;
 283
 284         if (client->ipv6) {
 285                 client->addr6.sin6_family = AF_INET6;
 286                 client->addr6.sin6_port = htons(client->port);
 287                 domain = AF_INET6;
 288                 addr = (struct sockaddr *) &client->addr6;
 289                 socklen = sizeof(client->addr6);
 290         } else {
 291                 client->addr.sin_family = AF_INET;
 292                 client->addr.sin_port = htons(client->port);
 293                 domain = AF_INET;
 294                 addr = (struct sockaddr *) &client->addr;
 295                 socklen = sizeof(client->addr);
 296         }
 297
 298         fd = socket(domain, SOCK_STREAM, 0);
 299         if (fd < 0) {
 300                 int ret = -errno;
 301
 302                 log_err("fio: socket: %s\n", strerror(errno));
 303                 return ret;
 304         }
 305
 306         if (connect(fd, addr, socklen) < 0) {
 307                 int ret = -errno;
 308
 309                 log_err("fio: connect: %s\n", strerror(errno));
 310                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 311                                                                 client->port);
 312                 close(fd);
 313                 return ret;
 314         }
 315
 316         return fd;
 317 }
 318
 319 static int fio_client_connect_sock(struct fio_client *client)
 320 {
 321         struct sockaddr_un *addr = &client->addr_un;
 322         fio_socklen_t len;
 323         int fd;
 324
 325         memset(addr, 0, sizeof(*addr));
 326         addr->sun_family = AF_UNIX;
 327         strcpy(addr->sun_path, client->hostname);
 328
 329         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 330         if (fd < 0) {
 331                 int ret = -errno;
 332
 333                 log_err("fio: socket: %s\n", strerror(errno));
 334                 return ret;
 335         }
 336
 337         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 338         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 339                 int ret = -errno;
 340
 341                 log_err("fio: connect; %s\n", strerror(errno));
 342                 close(fd);
 343                 return ret;
 344         }
 345
 346         return fd;
 347 }
 348
 349 int fio_client_connect(struct fio_client *client)
 350 {
 351         int fd;
 352
 353         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 354
 355         if (client->is_sock)
 356                 fd = fio_client_connect_sock(client);
 357         else
 358                 fd = fio_client_connect_ip(client);
 359
 360         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 361
 362         if (fd < 0)
 363                 return fd;
 364
 365         client->fd = fd;
 366         fio_client_add_hash(client);
 367         client->state = Client_connected;
 368
 369         probe_client(client);
 370         return 0;
 371 }
 372
 373 void fio_client_terminate(struct fio_client *client)
 374 {
 375         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0, NULL);
 376 }
 377
 378 void fio_clients_terminate(void)
 379 {
 380         struct flist_head *entry;
 381         struct fio_client *client;
 382
 383         dprint(FD_NET, "client: terminate clients\n");
 384
 385         flist_for_each(entry, &client_list) {
 386                 client = flist_entry(entry, struct fio_client, list);
 387                 fio_client_terminate(client);
 388         }
 389 }
 390
 391 static void sig_int(int sig)
 392 {
 393         dprint(FD_NET, "client: got signal %d\n", sig);
 394         fio_clients_terminate();
 395 }
 396
 397 static void client_signal_handler(void)
 398 {
 399         struct sigaction act;
 400
 401         memset(&act, 0, sizeof(act));
 402         act.sa_handler = sig_int;
 403         act.sa_flags = SA_RESTART;
 404         sigaction(SIGINT, &act, NULL);
 405
 406         memset(&act, 0, sizeof(act));
 407         act.sa_handler = sig_int;
 408         act.sa_flags = SA_RESTART;
 409         sigaction(SIGTERM, &act, NULL);
 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         clp->client_type = __cpu_to_le16(client->type);
 455         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
 456         free(pdu);
 457         return ret;
 458 }
 459
 460 int fio_clients_connect(void)
 461 {
 462         struct fio_client *client;
 463         struct flist_head *entry, *tmp;
 464         int ret;
 465
 466 #ifdef WIN32
 467         WSADATA wsd;
 468         WSAStartup(MAKEWORD(2,2), &wsd);
 469 #endif
 470
 471         dprint(FD_NET, "client: connect all\n");
 472
 473         client_signal_handler();
 474
 475         flist_for_each_safe(entry, tmp, &client_list) {
 476                 client = flist_entry(entry, struct fio_client, list);
 477
 478                 ret = fio_client_connect(client);
 479                 if (ret) {
 480                         remove_client(client);
 481                         continue;
 482                 }
 483
 484                 if (client->argc > 1)
 485                         send_client_cmd_line(client);
 486         }
 487
 488         return !nr_clients;
 489 }
 490
 491 int fio_start_client(struct fio_client *client)
 492 {
 493         dprint(FD_NET, "client: start %s\n", client->hostname);
 494         return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
 495 }
 496
 497 int fio_start_all_clients(void)
 498 {
 499         struct fio_client *client;
 500         struct flist_head *entry, *tmp;
 501         int ret;
 502
 503         dprint(FD_NET, "client: start all\n");
 504
 505         flist_for_each_safe(entry, tmp, &client_list) {
 506                 client = flist_entry(entry, struct fio_client, list);
 507
 508                 ret = fio_start_client(client);
 509                 if (ret) {
 510                         remove_client(client);
 511                         continue;
 512                 }
 513         }
 514
 515         return flist_empty(&client_list);
 516 }
 517
 518 /*
 519  * Send file contents to server backend. We could use sendfile(), but to remain
 520  * more portable lets just read/write the darn thing.
 521  */
 522 static int __fio_client_send_ini(struct fio_client *client, const char *filename)
 523 {
 524         struct cmd_job_pdu *pdu;
 525         size_t p_size;
 526         struct stat sb;
 527         char *p;
 528         void *buf;
 529         off_t len;
 530         int fd, ret;
 531
 532         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 533
 534         fd = open(filename, O_RDONLY);
 535         if (fd < 0) {
 536                 int ret = -errno;
 537
 538                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 539                 return ret;
 540         }
 541
 542         if (fstat(fd, &sb) < 0) {
 543                 int ret = -errno;
 544
 545                 log_err("fio: job file stat: %s\n", strerror(errno));
 546                 close(fd);
 547                 return ret;
 548         }
 549
 550         p_size = sb.st_size + sizeof(*pdu);
 551         pdu = malloc(p_size);
 552         buf = pdu->buf;
 553
 554         len = sb.st_size;
 555         p = buf;
 556         do {
 557                 ret = read(fd, p, len);
 558                 if (ret > 0) {
 559                         len -= ret;
 560                         if (!len)
 561                                 break;
 562                         p += ret;
 563                         continue;
 564                 } else if (!ret)
 565                         break;
 566                 else if (errno == EAGAIN || errno == EINTR)
 567                         continue;
 568         } while (1);
 569
 570         if (len) {
 571                 log_err("fio: failed reading job file %s\n", filename);
 572                 close(fd);
 573                 free(buf);
 574                 return 1;
 575         }
 576
 577         pdu->buf_len = __cpu_to_le32(sb.st_size);
 578         pdu->client_type = cpu_to_le32(client->type);
 579
 580         client->sent_job = 1;
 581         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, 0);
 582         free(pdu);
 583         close(fd);
 584         return ret;
 585 }
 586
 587 int fio_client_send_ini(struct fio_client *client, const char *filename)
 588 {
 589         int ret;
 590
 591         ret = __fio_client_send_ini(client, filename);
 592         if (!ret)
 593                 client->sent_job = 1;
 594
 595         return ret;
 596 }
 597
 598 int fio_clients_send_ini(const char *filename)
 599 {
 600         struct fio_client *client;
 601         struct flist_head *entry, *tmp;
 602
 603         flist_for_each_safe(entry, tmp, &client_list) {
 604                 client = flist_entry(entry, struct fio_client, list);
 605
 606                 if (fio_client_send_ini(client, filename))
 607                         remove_client(client);
 608         }
 609
 610         return !nr_clients;
 611 }
 612
 613 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 614 {
 615         dst->max_val    = le64_to_cpu(src->max_val);
 616         dst->min_val    = le64_to_cpu(src->min_val);
 617         dst->samples    = le64_to_cpu(src->samples);
 618
 619         /*
 620          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 621          */
 622         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 623         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 624 }
 625
 626 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 627 {
 628         int i, j;
 629
 630         dst->error      = le32_to_cpu(src->error);
 631         dst->groupid    = le32_to_cpu(src->groupid);
 632         dst->pid        = le32_to_cpu(src->pid);
 633         dst->members    = le32_to_cpu(src->members);
 634
 635         for (i = 0; i < 2; i++) {
 636                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 637                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 638                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 639                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 640         }
 641
 642         dst->usr_time           = le64_to_cpu(src->usr_time);
 643         dst->sys_time           = le64_to_cpu(src->sys_time);
 644         dst->ctx                = le64_to_cpu(src->ctx);
 645         dst->minf               = le64_to_cpu(src->minf);
 646         dst->majf               = le64_to_cpu(src->majf);
 647         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 648
 649         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 650                 fio_fp64_t *fps = &src->percentile_list[i];
 651                 fio_fp64_t *fpd = &dst->percentile_list[i];
 652
 653                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 654         }
 655
 656         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 657                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 658                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 659                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 660         }
 661
 662         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 663                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 664                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 665         }
 666
 667         for (i = 0; i < 2; i++)
 668                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 669                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 670
 671         for (i = 0; i < 3; i++) {
 672                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 673                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 674         }
 675
 676         dst->total_submit       = le64_to_cpu(src->total_submit);
 677         dst->total_complete     = le64_to_cpu(src->total_complete);
 678
 679         for (i = 0; i < 2; i++) {
 680                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 681                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 682         }
 683
 684         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 685         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 686         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 687         dst->first_error        = le32_to_cpu(src->first_error);
 688         dst->kb_base            = le32_to_cpu(src->kb_base);
 689 }
 690
 691 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 692 {
 693         int i;
 694
 695         for (i = 0; i < 2; i++) {
 696                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 697                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 698                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 699                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 700                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 701                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 702         }
 703
 704         dst->kb_base    = le32_to_cpu(src->kb_base);
 705         dst->groupid    = le32_to_cpu(src->groupid);
 706 }
 707
 708 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
 709 {
 710         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 711
 712         show_thread_status(&p->ts, &p->rs);
 713
 714         if (sum_stat_clients == 1)
 715                 return;
 716
 717         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 718         sum_group_stats(&client_gs, &p->rs);
 719
 720         client_ts.members++;
 721         client_ts.groupid = p->ts.groupid;
 722
 723         if (++sum_stat_nr == sum_stat_clients) {
 724                 strcpy(client_ts.name, "All clients");
 725                 show_thread_status(&client_ts, &client_gs);
 726         }
 727 }
 728
 729 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
 730 {
 731         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 732
 733         show_group_stats(gs);
 734 }
 735
 736 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
 737 {
 738         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
 739         const char *buf = (const char *) pdu->buf;
 740         const char *name;
 741         int fio_unused ret;
 742
 743         name = client->name ? client->name : client->hostname;
 744
 745         if (!client->skip_newline)
 746                 fprintf(f_out, "<%s> ", name);
 747         ret = fwrite(buf, pdu->buf_len, 1, f_out);
 748         fflush(f_out);
 749         client->skip_newline = strchr(buf, '\n') == NULL;
 750 }
 751
 752 static void convert_agg(struct disk_util_agg *agg)
 753 {
 754         int i;
 755
 756         for (i = 0; i < 2; i++) {
 757                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 758                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 759                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 760                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 761         }
 762
 763         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 764         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 765         agg->slavecount         = le32_to_cpu(agg->slavecount);
 766         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 767 }
 768
 769 static void convert_dus(struct disk_util_stat *dus)
 770 {
 771         int i;
 772
 773         for (i = 0; i < 2; i++) {
 774                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
 775                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
 776                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
 777                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
 778         }
 779
 780         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
 781         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
 782         dus->msec               = le64_to_cpu(dus->msec);
 783 }
 784
 785 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 786 {
 787         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 788
 789         if (!client->disk_stats_shown) {
 790                 client->disk_stats_shown = 1;
 791                 log_info("\nDisk stats (read/write):\n");
 792         }
 793
 794         print_disk_util(&du->dus, &du->agg, terse_output);
 795 }
 796
 797 static void convert_jobs_eta(struct jobs_eta *je)
 798 {
 799         int i;
 800
 801         je->nr_running          = le32_to_cpu(je->nr_running);
 802         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 803         je->nr_pending          = le32_to_cpu(je->nr_pending);
 804         je->files_open          = le32_to_cpu(je->files_open);
 805
 806         for (i = 0; i < 2; i++) {
 807                 je->m_rate[i]           = le32_to_cpu(je->m_rate[i]);
 808                 je->t_rate[i]           = le32_to_cpu(je->t_rate[i]);
 809                 je->m_iops[i]           = le32_to_cpu(je->m_iops[i]);
 810                 je->t_iops[i]           = le32_to_cpu(je->t_iops[i]);
 811                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 812                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 813         }
 814
 815         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 816         je->eta_sec             = le64_to_cpu(je->eta_sec);
 817         je->nr_threads          = le32_to_cpu(je->nr_threads);
 818 }
 819
 820 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 821 {
 822         int i;
 823
 824         dst->nr_running         += je->nr_running;
 825         dst->nr_ramp            += je->nr_ramp;
 826         dst->nr_pending         += je->nr_pending;
 827         dst->files_open         += je->files_open;
 828
 829         for (i = 0; i < 2; i++) {
 830                 dst->m_rate[i]  += je->m_rate[i];
 831                 dst->t_rate[i]  += je->t_rate[i];
 832                 dst->m_iops[i]  += je->m_iops[i];
 833                 dst->t_iops[i]  += je->t_iops[i];
 834                 dst->rate[i]    += je->rate[i];
 835                 dst->iops[i]    += je->iops[i];
 836         }
 837
 838         dst->elapsed_sec        += je->elapsed_sec;
 839
 840         if (je->eta_sec > dst->eta_sec)
 841                 dst->eta_sec = je->eta_sec;
 842
 843         dst->nr_threads         += je->nr_threads;
 844         /* we need to handle je->run_str too ... */
 845 }
 846
 847 void fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
 848 {
 849         if (!--eta->pending) {
 850                 eta_fn(&eta->eta);
 851                 free(eta);
 852         }
 853 }
 854
 855 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 856 {
 857         struct fio_net_int_cmd *icmd = NULL;
 858         struct flist_head *entry;
 859
 860         flist_for_each(entry, &client->cmd_list) {
 861                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
 862
 863                 if (cmd->tag == (uintptr_t) icmd)
 864                         break;
 865
 866                 icmd = NULL;
 867         }
 868
 869         if (!icmd) {
 870                 log_err("fio: client: unable to find matching tag\n");
 871                 return;
 872         }
 873
 874         flist_del(&icmd->list);
 875         cmd->tag = icmd->saved_tag;
 876         free(icmd);
 877 }
 878
 879 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 880 {
 881         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 882         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 883
 884         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 885
 886         assert(client->eta_in_flight == eta);
 887
 888         client->eta_in_flight = NULL;
 889         flist_del_init(&client->eta_list);
 890
 891         if (client->ops->jobs_eta)
 892                 client->ops->jobs_eta(client, je);
 893
 894         fio_client_sum_jobs_eta(&eta->eta, je);
 895         fio_client_dec_jobs_eta(eta, client->ops->eta);
 896 }
 897
 898 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 899 {
 900         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 901         const char *os, *arch;
 902         char bit[16];
 903
 904         os = fio_get_os_string(probe->os);
 905         if (!os)
 906                 os = "unknown";
 907
 908         arch = fio_get_arch_string(probe->arch);
 909         if (!arch)
 910                 os = "unknown";
 911
 912         sprintf(bit, "%d-bit", probe->bpp * 8);
 913
 914         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
 915                 probe->hostname, probe->bigendian, bit, os, arch,
 916                 probe->fio_major, probe->fio_minor, probe->fio_patch);
 917
 918         if (!client->name)
 919                 client->name = strdup((char *) probe->hostname);
 920 }
 921
 922 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
 923 {
 924         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 925
 926         client->state = Client_started;
 927         client->jobs = pdu->jobs;
 928 }
 929
 930 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
 931 {
 932         if (client->error)
 933                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
 934 }
 935
 936 static void convert_stop(struct fio_net_cmd *cmd)
 937 {
 938         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
 939
 940         pdu->error = le32_to_cpu(pdu->error);
 941 }
 942
 943 static void convert_text(struct fio_net_cmd *cmd)
 944 {
 945         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
 946
 947         pdu->level      = le32_to_cpu(pdu->level);
 948         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
 949         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
 950         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
 951 }
 952
 953 int fio_handle_client(struct fio_client *client)
 954 {
 955         struct client_ops *ops = client->ops;
 956         struct fio_net_cmd *cmd;
 957
 958         dprint(FD_NET, "client: handle %s\n", client->hostname);
 959
 960         cmd = fio_net_recv_cmd(client->fd);
 961         if (!cmd)
 962                 return 0;
 963
 964         dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
 965                 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
 966
 967         switch (cmd->opcode) {
 968         case FIO_NET_CMD_QUIT:
 969                 if (ops->quit)
 970                         ops->quit(client, cmd);
 971                 remove_client(client);
 972                 free(cmd);
 973                 break;
 974         case FIO_NET_CMD_TEXT:
 975                 convert_text(cmd);
 976                 ops->text(client, cmd);
 977                 free(cmd);
 978                 break;
 979         case FIO_NET_CMD_DU: {
 980                 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 981
 982                 convert_dus(&du->dus);
 983                 convert_agg(&du->agg);
 984
 985                 ops->disk_util(client, cmd);
 986                 free(cmd);
 987                 break;
 988                 }
 989         case FIO_NET_CMD_TS: {
 990                 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 991
 992                 convert_ts(&p->ts, &p->ts);
 993                 convert_gs(&p->rs, &p->rs);
 994
 995                 ops->thread_status(client, cmd);
 996                 free(cmd);
 997                 break;
 998                 }
 999         case FIO_NET_CMD_GS: {
1000                 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1001
1002                 convert_gs(gs, gs);
1003
1004                 ops->group_stats(client, cmd);
1005                 free(cmd);
1006                 break;
1007                 }
1008         case FIO_NET_CMD_ETA: {
1009                 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1010
1011                 remove_reply_cmd(client, cmd);
1012                 convert_jobs_eta(je);
1013                 handle_eta(client, cmd);
1014                 free(cmd);
1015                 break;
1016                 }
1017         case FIO_NET_CMD_PROBE:
1018                 remove_reply_cmd(client, cmd);
1019                 ops->probe(client, cmd);
1020                 free(cmd);
1021                 break;
1022         case FIO_NET_CMD_SERVER_START:
1023                 client->state = Client_running;
1024                 if (ops->job_start)
1025                         ops->job_start(client, cmd);
1026                 free(cmd);
1027                 break;
1028         case FIO_NET_CMD_START: {
1029                 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1030
1031                 pdu->jobs = le32_to_cpu(pdu->jobs);
1032                 ops->start(client, cmd);
1033                 free(cmd);
1034                 break;
1035                 }
1036         case FIO_NET_CMD_STOP: {
1037                 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1038
1039                 convert_stop(cmd);
1040                 client->state = Client_stopped;
1041                 client->error = pdu->error;
1042                 ops->stop(client, cmd);
1043                 free(cmd);
1044                 break;
1045                 }
1046         case FIO_NET_CMD_ADD_JOB:
1047                 if (ops->add_job)
1048                         ops->add_job(client, cmd);
1049                 free(cmd);
1050                 break;
1051         default:
1052                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1053                 free(cmd);
1054                 break;
1055         }
1056
1057         return 1;
1058 }
1059
1060 static void request_client_etas(struct client_ops *ops)
1061 {
1062         struct fio_client *client;
1063         struct flist_head *entry;
1064         struct client_eta *eta;
1065         int skipped = 0;
1066
1067         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1068
1069         eta = malloc(sizeof(*eta));
1070         memset(&eta->eta, 0, sizeof(eta->eta));
1071         eta->pending = nr_clients;
1072
1073         flist_for_each(entry, &client_list) {
1074                 client = flist_entry(entry, struct fio_client, list);
1075
1076                 if (!flist_empty(&client->eta_list)) {
1077                         skipped++;
1078                         continue;
1079                 }
1080                 if (client->state != Client_running)
1081                         continue;
1082
1083                 assert(!client->eta_in_flight);
1084                 flist_add_tail(&client->eta_list, &eta_list);
1085                 client->eta_in_flight = eta;
1086                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1087                                         (uintptr_t) eta, &client->cmd_list);
1088         }
1089
1090         while (skipped--)
1091                 fio_client_dec_jobs_eta(eta, ops->eta);
1092
1093         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1094 }
1095
1096 static int client_check_cmd_timeout(struct fio_client *client,
1097                                     struct timeval *now)
1098 {
1099         struct fio_net_int_cmd *cmd;
1100         struct flist_head *entry, *tmp;
1101         int ret = 0;
1102
1103         flist_for_each_safe(entry, tmp, &client->cmd_list) {
1104                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
1105
1106                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
1107                         continue;
1108
1109                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1110                                                 fio_server_op(cmd->cmd.opcode));
1111                 flist_del(&cmd->list);
1112                 free(cmd);
1113                 ret = 1;
1114         }
1115
1116         return flist_empty(&client->cmd_list) && ret;
1117 }
1118
1119 static int fio_check_clients_timed_out(void)
1120 {
1121         struct fio_client *client;
1122         struct flist_head *entry, *tmp;
1123         struct timeval tv;
1124         int ret = 0;
1125
1126         gettimeofday(&tv, NULL);
1127
1128         flist_for_each_safe(entry, tmp, &client_list) {
1129                 client = flist_entry(entry, struct fio_client, list);
1130
1131                 if (flist_empty(&client->cmd_list))
1132                         continue;
1133
1134                 if (!client_check_cmd_timeout(client, &tv))
1135                         continue;
1136
1137                 if (client->ops->timed_out)
1138                         client->ops->timed_out(client);
1139                 else
1140                         log_err("fio: client %s timed out\n", client->hostname);
1141
1142                 remove_client(client);
1143                 ret = 1;
1144         }
1145
1146         return ret;
1147 }
1148
1149 int fio_handle_clients(struct client_ops *ops)
1150 {
1151         struct pollfd *pfds;
1152         int i, ret = 0, retval = 0;
1153
1154         gettimeofday(&eta_tv, NULL);
1155
1156         pfds = malloc(nr_clients * sizeof(struct pollfd));
1157
1158         sum_stat_clients = nr_clients;
1159         init_thread_stat(&client_ts);
1160         init_group_run_stat(&client_gs);
1161
1162         while (!exit_backend && nr_clients) {
1163                 struct flist_head *entry, *tmp;
1164                 struct fio_client *client;
1165
1166                 i = 0;
1167                 flist_for_each_safe(entry, tmp, &client_list) {
1168                         client = flist_entry(entry, struct fio_client, list);
1169
1170                         if (!client->sent_job && !client->ops->stay_connected &&
1171                             flist_empty(&client->cmd_list)) {
1172                                 remove_client(client);
1173                                 continue;
1174                         }
1175
1176                         pfds[i].fd = client->fd;
1177                         pfds[i].events = POLLIN;
1178                         i++;
1179                 }
1180
1181                 if (!nr_clients)
1182                         break;
1183
1184                 assert(i == nr_clients);
1185
1186                 do {
1187                         struct timeval tv;
1188
1189                         gettimeofday(&tv, NULL);
1190                         if (mtime_since(&eta_tv, &tv) >= ops->eta_msec) {
1191                                 request_client_etas(ops);
1192                                 memcpy(&eta_tv, &tv, sizeof(tv));
1193
1194                                 if (fio_check_clients_timed_out())
1195                                         break;
1196                         }
1197
1198                         ret = poll(pfds, nr_clients, 100);
1199                         if (ret < 0) {
1200                                 if (errno == EINTR)
1201                                         continue;
1202                                 log_err("fio: poll clients: %s\n", strerror(errno));
1203                                 break;
1204                         } else if (!ret)
1205                                 continue;
1206                 } while (ret <= 0);
1207
1208                 for (i = 0; i < nr_clients; i++) {
1209                         if (!(pfds[i].revents & POLLIN))
1210                                 continue;
1211
1212                         client = find_client_by_fd(pfds[i].fd);
1213                         if (!client) {
1214                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1215                                 continue;
1216                         }
1217                         if (!fio_handle_client(client)) {
1218                                 log_info("client: host=%s disconnected\n",
1219                                                 client->hostname);
1220                                 remove_client(client);
1221                                 retval = 1;
1222                         } else if (client->error)
1223                                 retval = 1;
1224                         fio_put_client(client);
1225                 }
1226         }
1227
1228         free(pfds);
1229         return retval;
1230 }