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