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 extern void (*update_thread_status)(char *status_message);
  25
  26 struct client_eta {
  27         struct jobs_eta eta;
  28         unsigned int pending;
  29 };
  30
  31 struct fio_client {
  32         struct flist_head list;
  33         struct flist_head hash_list;
  34         struct flist_head arg_list;
  35         union {
  36                 struct sockaddr_in addr;
  37                 struct sockaddr_in6 addr6;
  38                 struct sockaddr_un addr_un;
  39         };
  40         char *hostname;
  41         int port;
  42         int fd;
  43
  44         char *name;
  45
  46         int state;
  47
  48         int skip_newline;
  49         int is_sock;
  50         int disk_stats_shown;
  51         unsigned int jobs;
  52         int error;
  53         int ipv6;
  54         int sent_job;
  55
  56         struct flist_head eta_list;
  57         struct client_eta *eta_in_flight;
  58
  59         struct flist_head cmd_list;
  60
  61         uint16_t argc;
  62         char **argv;
  63 };
  64
  65 static void fio_client_text_op(struct fio_client *client,
  66                 FILE *f, __u16 pdu_len, const char *buf)
  67 {
  68         const char *name;
  69         int fio_unused ret;
  70
  71         name = client->name ? client->name : client->hostname;
  72
  73         if (!client->skip_newline)
  74                 fprintf(f, "<%s> ", name);
  75         ret = fwrite(buf, pdu_len, 1, f);
  76         fflush(f);
  77         client->skip_newline = strchr(buf, '\n') == NULL;
  78 }
  79
  80 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
  81 static void handle_ts(struct fio_net_cmd *cmd);
  82 static void handle_gs(struct fio_net_cmd *cmd);
  83 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd);
  84 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
  85
  86 struct client_ops fio_client_ops = {
  87         fio_client_text_op,
  88         handle_du,
  89         handle_ts,
  90         handle_gs,
  91         handle_eta,
  92         handle_probe,
  93         NULL, /* status display, if NULL, printf is used */
  94 };
  95
  96 static struct timeval eta_tv;
  97
  98 enum {
  99         Client_created          = 0,
 100         Client_connected        = 1,
 101         Client_started          = 2,
 102         Client_running          = 3,
 103         Client_stopped          = 4,
 104         Client_exited           = 5,
 105 };
 106
 107 static FLIST_HEAD(client_list);
 108 static FLIST_HEAD(eta_list);
 109
 110 static FLIST_HEAD(arg_list);
 111
 112 static struct thread_stat client_ts;
 113 static struct group_run_stats client_gs;
 114 static int sum_stat_clients;
 115 static int sum_stat_nr;
 116
 117 #define FIO_CLIENT_HASH_BITS    7
 118 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
 119 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
 120 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
 121
 122 static int handle_client(struct fio_client *client, struct client_ops *ops);
 123 static void dec_jobs_eta(struct client_eta *eta);
 124
 125 static void fio_client_add_hash(struct fio_client *client)
 126 {
 127         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
 128
 129         bucket &= FIO_CLIENT_HASH_MASK;
 130         flist_add(&client->hash_list, &client_hash[bucket]);
 131 }
 132
 133 static void fio_client_remove_hash(struct fio_client *client)
 134 {
 135         if (!flist_empty(&client->hash_list))
 136                 flist_del_init(&client->hash_list);
 137 }
 138
 139 static void fio_init fio_client_hash_init(void)
 140 {
 141         int i;
 142
 143         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
 144                 INIT_FLIST_HEAD(&client_hash[i]);
 145 }
 146
 147 static struct fio_client *find_client_by_fd(int fd)
 148 {
 149         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
 150         struct fio_client *client;
 151         struct flist_head *entry;
 152
 153         flist_for_each(entry, &client_hash[bucket]) {
 154                 client = flist_entry(entry, struct fio_client, hash_list);
 155
 156                 if (client->fd == fd)
 157                         return client;
 158         }
 159
 160         return NULL;
 161 }
 162
 163 static void remove_client(struct fio_client *client)
 164 {
 165         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 166         flist_del(&client->list);
 167
 168         fio_client_remove_hash(client);
 169
 170         if (!flist_empty(&client->eta_list)) {
 171                 flist_del_init(&client->eta_list);
 172                 dec_jobs_eta(client->eta_in_flight);
 173         }
 174
 175         free(client->hostname);
 176         if (client->argv)
 177                 free(client->argv);
 178         if (client->name)
 179                 free(client->name);
 180
 181         free(client);
 182         nr_clients--;
 183         sum_stat_clients--;
 184 }
 185
 186 static void __fio_client_add_cmd_option(struct fio_client *client,
 187                                         const char *opt)
 188 {
 189         int index;
 190
 191         index = client->argc++;
 192         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 193         client->argv[index] = strdup(opt);
 194         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 195 }
 196
 197 void fio_client_add_cmd_option(void *cookie, const char *opt)
 198 {
 199         struct fio_client *client = cookie;
 200         struct flist_head *entry;
 201
 202         if (!client || !opt)
 203                 return;
 204
 205         __fio_client_add_cmd_option(client, opt);
 206
 207         /*
 208          * Duplicate arguments to shared client group
 209          */
 210         flist_for_each(entry, &arg_list) {
 211                 client = flist_entry(entry, struct fio_client, arg_list);
 212
 213                 __fio_client_add_cmd_option(client, opt);
 214         }
 215 }
 216
 217 int fio_client_add(const char *hostname, void **cookie)
 218 {
 219         struct fio_client *existing = *cookie;
 220         struct fio_client *client;
 221
 222         if (existing) {
 223                 /*
 224                  * We always add our "exec" name as the option, hence 1
 225                  * means empty.
 226                  */
 227                 if (existing->argc == 1)
 228                         flist_add_tail(&existing->arg_list, &arg_list);
 229                 else {
 230                         while (!flist_empty(&arg_list))
 231                                 flist_del_init(arg_list.next);
 232                 }
 233         }
 234
 235         client = malloc(sizeof(*client));
 236         memset(client, 0, sizeof(*client));
 237
 238         INIT_FLIST_HEAD(&client->list);
 239         INIT_FLIST_HEAD(&client->hash_list);
 240         INIT_FLIST_HEAD(&client->arg_list);
 241         INIT_FLIST_HEAD(&client->eta_list);
 242         INIT_FLIST_HEAD(&client->cmd_list);
 243
 244         if (fio_server_parse_string(hostname, &client->hostname,
 245                                         &client->is_sock, &client->port,
 246                                         &client->addr.sin_addr,
 247                                         &client->addr6.sin6_addr,
 248                                         &client->ipv6))
 249                 return -1;
 250
 251         client->fd = -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_net_cmd *cmd)
 639 {
 640         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 641
 642         convert_ts(&p->ts, &p->ts);
 643         convert_gs(&p->rs, &p->rs);
 644
 645         show_thread_status(&p->ts, &p->rs);
 646
 647         if (sum_stat_clients == 1)
 648                 return;
 649
 650         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 651         sum_group_stats(&client_gs, &p->rs);
 652
 653         client_ts.members++;
 654         client_ts.groupid = p->ts.groupid;
 655
 656         if (++sum_stat_nr == sum_stat_clients) {
 657                 strcpy(client_ts.name, "All clients");
 658                 show_thread_status(&client_ts, &client_gs);
 659         }
 660 }
 661
 662 static void handle_gs(struct fio_net_cmd *cmd)
 663 {
 664         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 665
 666         convert_gs(gs, gs);
 667         show_group_stats(gs);
 668 }
 669
 670 static void convert_agg(struct disk_util_agg *agg)
 671 {
 672         int i;
 673
 674         for (i = 0; i < 2; i++) {
 675                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 676                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 677                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 678                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 679         }
 680
 681         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 682         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 683         agg->slavecount         = le32_to_cpu(agg->slavecount);
 684         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 685 }
 686
 687 static void convert_dus(struct disk_util_stat *dus)
 688 {
 689         int i;
 690
 691         for (i = 0; i < 2; i++) {
 692                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
 693                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
 694                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
 695                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
 696         }
 697
 698         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
 699         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
 700         dus->msec               = le64_to_cpu(dus->msec);
 701 }
 702
 703 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 704 {
 705         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 706
 707         convert_dus(&du->dus);
 708         convert_agg(&du->agg);
 709
 710         if (!client->disk_stats_shown) {
 711                 client->disk_stats_shown = 1;
 712                 log_info("\nDisk stats (read/write):\n");
 713         }
 714
 715         print_disk_util(&du->dus, &du->agg, terse_output);
 716 }
 717
 718 static void convert_jobs_eta(struct jobs_eta *je)
 719 {
 720         int i;
 721
 722         je->nr_running          = le32_to_cpu(je->nr_running);
 723         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 724         je->nr_pending          = le32_to_cpu(je->nr_pending);
 725         je->files_open          = le32_to_cpu(je->files_open);
 726         je->m_rate              = le32_to_cpu(je->m_rate);
 727         je->t_rate              = le32_to_cpu(je->t_rate);
 728         je->m_iops              = le32_to_cpu(je->m_iops);
 729         je->t_iops              = le32_to_cpu(je->t_iops);
 730
 731         for (i = 0; i < 2; i++) {
 732                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 733                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 734         }
 735
 736         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 737         je->eta_sec             = le64_to_cpu(je->eta_sec);
 738 }
 739
 740 static void sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 741 {
 742         int i;
 743
 744         dst->nr_running         += je->nr_running;
 745         dst->nr_ramp            += je->nr_ramp;
 746         dst->nr_pending         += je->nr_pending;
 747         dst->files_open         += je->files_open;
 748         dst->m_rate             += je->m_rate;
 749         dst->t_rate             += je->t_rate;
 750         dst->m_iops             += je->m_iops;
 751         dst->t_iops             += je->t_iops;
 752
 753         for (i = 0; i < 2; i++) {
 754                 dst->rate[i]    += je->rate[i];
 755                 dst->iops[i]    += je->iops[i];
 756         }
 757
 758         dst->elapsed_sec        += je->elapsed_sec;
 759
 760         if (je->eta_sec > dst->eta_sec)
 761                 dst->eta_sec = je->eta_sec;
 762 }
 763
 764 static void dec_jobs_eta(struct client_eta *eta)
 765 {
 766         if (!--eta->pending) {
 767                 display_thread_status(&eta->eta);
 768                 free(eta);
 769         }
 770 }
 771
 772 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 773 {
 774         struct fio_net_int_cmd *icmd = NULL;
 775         struct flist_head *entry;
 776
 777         flist_for_each(entry, &client->cmd_list) {
 778                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
 779
 780                 if (cmd->tag == (uintptr_t) icmd)
 781                         break;
 782
 783                 icmd = NULL;
 784         }
 785
 786         if (!icmd) {
 787                 log_err("fio: client: unable to find matching tag\n");
 788                 return;
 789         }
 790
 791         flist_del(&icmd->list);
 792         cmd->tag = icmd->saved_tag;
 793         free(icmd);
 794 }
 795
 796 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 797 {
 798         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 799         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 800
 801         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 802
 803         assert(client->eta_in_flight == eta);
 804
 805         client->eta_in_flight = NULL;
 806         flist_del_init(&client->eta_list);
 807
 808         convert_jobs_eta(je);
 809         sum_jobs_eta(&eta->eta, je);
 810         dec_jobs_eta(eta);
 811 }
 812
 813 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 814 {
 815         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 816         const char *os, *arch;
 817         char bit[16];
 818
 819         os = fio_get_os_string(probe->os);
 820         if (!os)
 821                 os = "unknown";
 822
 823         arch = fio_get_arch_string(probe->arch);
 824         if (!arch)
 825                 os = "unknown";
 826
 827         sprintf(bit, "%d-bit", probe->bpp * 8);
 828
 829         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
 830                 probe->hostname, probe->bigendian, bit, os, arch,
 831                 probe->fio_major, probe->fio_minor, probe->fio_patch);
 832
 833         if (!client->name)
 834                 client->name = strdup((char *) probe->hostname);
 835 }
 836
 837 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
 838 {
 839         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 840
 841         client->state = Client_started;
 842         client->jobs = le32_to_cpu(pdu->jobs);
 843 }
 844
 845 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
 846 {
 847         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
 848
 849         client->state = Client_stopped;
 850         client->error = le32_to_cpu(pdu->error);
 851
 852         if (client->error)
 853                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
 854 }
 855
 856 static int handle_client(struct fio_client *client, struct client_ops *ops)
 857 {
 858         struct fio_net_cmd *cmd;
 859
 860         dprint(FD_NET, "client: handle %s\n", client->hostname);
 861
 862         cmd = fio_net_recv_cmd(client->fd);
 863         if (!cmd)
 864                 return 0;
 865
 866         dprint(FD_NET, "client: got cmd op %s from %s\n",
 867                                 fio_server_op(cmd->opcode), client->hostname);
 868
 869         switch (cmd->opcode) {
 870         case FIO_NET_CMD_QUIT:
 871                 remove_client(client);
 872                 free(cmd);
 873                 break;
 874         case FIO_NET_CMD_TEXT: {
 875                 const char *buf = (const char *) cmd->payload;
 876                 ops->text_op(client, f_out, cmd->pdu_len, buf);
 877                 free(cmd);
 878                 break;
 879                 }
 880         case FIO_NET_CMD_DU:
 881                 ops->disk_util(client, cmd);
 882                 free(cmd);
 883                 break;
 884         case FIO_NET_CMD_TS:
 885                 ops->thread_status(cmd);
 886                 free(cmd);
 887                 break;
 888         case FIO_NET_CMD_GS:
 889                 ops->group_stats(cmd);
 890                 free(cmd);
 891                 break;
 892         case FIO_NET_CMD_ETA:
 893                 remove_reply_cmd(client, cmd);
 894                 ops->eta(client, cmd);
 895                 free(cmd);
 896                 break;
 897         case FIO_NET_CMD_PROBE:
 898                 remove_reply_cmd(client, cmd);
 899                 ops->probe(client, cmd);
 900                 free(cmd);
 901                 break;
 902         case FIO_NET_CMD_RUN:
 903                 client->state = Client_running;
 904                 free(cmd);
 905                 break;
 906         case FIO_NET_CMD_START:
 907                 handle_start(client, cmd);
 908                 free(cmd);
 909                 break;
 910         case FIO_NET_CMD_STOP:
 911                 handle_stop(client, cmd);
 912                 free(cmd);
 913                 break;
 914         default:
 915                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
 916                 free(cmd);
 917                 break;
 918         }
 919
 920         return 1;
 921 }
 922
 923 static void request_client_etas(void)
 924 {
 925         struct fio_client *client;
 926         struct flist_head *entry;
 927         struct client_eta *eta;
 928         int skipped = 0;
 929
 930         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
 931
 932         eta = malloc(sizeof(*eta));
 933         memset(&eta->eta, 0, sizeof(eta->eta));
 934         eta->pending = nr_clients;
 935
 936         flist_for_each(entry, &client_list) {
 937                 client = flist_entry(entry, struct fio_client, list);
 938
 939                 if (!flist_empty(&client->eta_list)) {
 940                         skipped++;
 941                         continue;
 942                 }
 943                 if (client->state != Client_running)
 944                         continue;
 945
 946                 assert(!client->eta_in_flight);
 947                 flist_add_tail(&client->eta_list, &eta_list);
 948                 client->eta_in_flight = eta;
 949                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
 950                                         (uintptr_t) eta, &client->cmd_list);
 951         }
 952
 953         while (skipped--)
 954                 dec_jobs_eta(eta);
 955
 956         dprint(FD_NET, "client: requested eta tag %p\n", eta);
 957 }
 958
 959 static int client_check_cmd_timeout(struct fio_client *client,
 960                                     struct timeval *now)
 961 {
 962         struct fio_net_int_cmd *cmd;
 963         struct flist_head *entry, *tmp;
 964         int ret = 0;
 965
 966         flist_for_each_safe(entry, tmp, &client->cmd_list) {
 967                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
 968
 969                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
 970                         continue;
 971
 972                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
 973                                                 fio_server_op(cmd->cmd.opcode));
 974                 flist_del(&cmd->list);
 975                 free(cmd);
 976                 ret = 1;
 977         }
 978
 979         return flist_empty(&client->cmd_list) && ret;
 980 }
 981
 982 static int fio_client_timed_out(void)
 983 {
 984         struct fio_client *client;
 985         struct flist_head *entry, *tmp;
 986         struct timeval tv;
 987         int ret = 0;
 988
 989         gettimeofday(&tv, NULL);
 990
 991         flist_for_each_safe(entry, tmp, &client_list) {
 992                 client = flist_entry(entry, struct fio_client, list);
 993
 994                 if (flist_empty(&client->cmd_list))
 995                         continue;
 996
 997                 if (!client_check_cmd_timeout(client, &tv))
 998                         continue;
 999
1000                 log_err("fio: client %s timed out\n", client->hostname);
1001                 remove_client(client);
1002                 ret = 1;
1003         }
1004
1005         return ret;
1006 }
1007
1008 int fio_handle_clients(struct client_ops *ops)
1009 {
1010         struct pollfd *pfds;
1011         int i, ret = 0, retval = 0;
1012
1013         gettimeofday(&eta_tv, NULL);
1014
1015         pfds = malloc(nr_clients * sizeof(struct pollfd));
1016
1017         sum_stat_clients = nr_clients;
1018         init_thread_stat(&client_ts);
1019         init_group_run_stat(&client_gs);
1020
1021         /* Used by eta.c:display_thread_status() */
1022         update_thread_status = ops->thread_status_display;
1023
1024         while (!exit_backend && nr_clients) {
1025                 struct flist_head *entry, *tmp;
1026                 struct fio_client *client;
1027
1028                 i = 0;
1029                 flist_for_each_safe(entry, tmp, &client_list) {
1030                         client = flist_entry(entry, struct fio_client, list);
1031
1032                         if (!client->sent_job &&
1033                             flist_empty(&client->cmd_list)) {
1034                                 remove_client(client);
1035                                 continue;
1036                         }
1037
1038                         pfds[i].fd = client->fd;
1039                         pfds[i].events = POLLIN;
1040                         i++;
1041                 }
1042
1043                 if (!nr_clients)
1044                         break;
1045
1046                 assert(i == nr_clients);
1047
1048                 do {
1049                         struct timeval tv;
1050
1051                         gettimeofday(&tv, NULL);
1052                         if (mtime_since(&eta_tv, &tv) >= 900) {
1053                                 request_client_etas();
1054                                 memcpy(&eta_tv, &tv, sizeof(tv));
1055
1056                                 if (fio_client_timed_out())
1057                                         break;
1058                         }
1059
1060                         ret = poll(pfds, nr_clients, 100);
1061                         if (ret < 0) {
1062                                 if (errno == EINTR)
1063                                         continue;
1064                                 log_err("fio: poll clients: %s\n", strerror(errno));
1065                                 break;
1066                         } else if (!ret)
1067                                 continue;
1068                 } while (ret <= 0);
1069
1070                 for (i = 0; i < nr_clients; i++) {
1071                         if (!(pfds[i].revents & POLLIN))
1072                                 continue;
1073
1074                         client = find_client_by_fd(pfds[i].fd);
1075                         if (!client) {
1076                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1077                                 continue;
1078                         }
1079                         if (!handle_client(client, ops)) {
1080                                 log_info("client: host=%s disconnected\n",
1081                                                 client->hostname);
1082                                 remove_client(client);
1083                                 retval = 1;
1084                         } else if (client->error)
1085                                 retval = 1;
1086                 }
1087         }
1088
1089         free(pfds);
1090         return retval;
1091 }