client.c

   1 #include <stdio.h>
   2 #include <stdlib.h>
   3 #include <unistd.h>
   4 #include <limits.h>
   5 #include <errno.h>
   6 #include <fcntl.h>
   7 #include <sys/poll.h>
   8 #include <sys/types.h>
   9 #include <sys/stat.h>
  10 #include <sys/wait.h>
  11 #include <sys/socket.h>
  12 #include <sys/un.h>
  13 #include <netinet/in.h>
  14 #include <arpa/inet.h>
  15 #include <netdb.h>
  16 #include <signal.h>
  17
  18 #include "fio.h"
  19 #include "server.h"
  20 #include "flist.h"
  21 #include "hash.h"
  22
  23 struct client_eta {
  24         struct jobs_eta eta;
  25         unsigned int pending;
  26 };
  27
  28 struct fio_client {
  29         struct flist_head list;
  30         struct flist_head hash_list;
  31         struct flist_head arg_list;
  32         struct sockaddr_in addr;
  33         struct sockaddr_un addr_un;
  34         char *hostname;
  35         int port;
  36         int fd;
  37
  38         char *name;
  39
  40         int state;
  41
  42         int skip_newline;
  43         int is_sock;
  44
  45         struct flist_head eta_list;
  46         struct client_eta *eta_in_flight;
  47
  48         struct flist_head cmd_list;
  49
  50         uint16_t argc;
  51         char **argv;
  52 };
  53
  54 static struct timeval eta_tv;
  55
  56 enum {
  57         Client_created          = 0,
  58         Client_connected        = 1,
  59         Client_started          = 2,
  60         Client_stopped          = 3,
  61         Client_exited           = 4,
  62 };
  63
  64 static FLIST_HEAD(client_list);
  65 static FLIST_HEAD(eta_list);
  66
  67 static FLIST_HEAD(arg_list);
  68
  69 static struct thread_stat client_ts;
  70 static struct group_run_stats client_gs;
  71 static int sum_stat_clients;
  72 static int sum_stat_nr;
  73
  74 #define FIO_CLIENT_HASH_BITS    7
  75 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  76 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  77 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  78
  79 static int handle_client(struct fio_client *client);
  80 static void dec_jobs_eta(struct client_eta *eta);
  81
  82 static void fio_client_add_hash(struct fio_client *client)
  83 {
  84         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  85
  86         bucket &= FIO_CLIENT_HASH_MASK;
  87         flist_add(&client->hash_list, &client_hash[bucket]);
  88 }
  89
  90 static void fio_client_remove_hash(struct fio_client *client)
  91 {
  92         if (!flist_empty(&client->hash_list))
  93                 flist_del_init(&client->hash_list);
  94 }
  95
  96 static void fio_init fio_client_hash_init(void)
  97 {
  98         int i;
  99
 100         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
 101                 INIT_FLIST_HEAD(&client_hash[i]);
 102 }
 103
 104 static struct fio_client *find_client_by_fd(int fd)
 105 {
 106         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
 107         struct fio_client *client;
 108         struct flist_head *entry;
 109
 110         flist_for_each(entry, &client_hash[bucket]) {
 111                 client = flist_entry(entry, struct fio_client, hash_list);
 112
 113                 if (client->fd == fd)
 114                         return client;
 115         }
 116
 117         return NULL;
 118 }
 119
 120 static void remove_client(struct fio_client *client)
 121 {
 122         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 123         flist_del(&client->list);
 124
 125         fio_client_remove_hash(client);
 126
 127         if (!flist_empty(&client->eta_list)) {
 128                 flist_del_init(&client->eta_list);
 129                 dec_jobs_eta(client->eta_in_flight);
 130         }
 131
 132         free(client->hostname);
 133         if (client->argv)
 134                 free(client->argv);
 135         if (client->name)
 136                 free(client->name);
 137
 138         free(client);
 139         nr_clients--;
 140         sum_stat_clients--;
 141 }
 142
 143 static void __fio_client_add_cmd_option(struct fio_client *client,
 144                                         const char *opt)
 145 {
 146         int index;
 147
 148         index = client->argc++;
 149         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 150         client->argv[index] = strdup(opt);
 151         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 152 }
 153
 154 void fio_client_add_cmd_option(void *cookie, const char *opt)
 155 {
 156         struct fio_client *client = cookie;
 157         struct flist_head *entry;
 158
 159         if (!client || !opt)
 160                 return;
 161
 162         __fio_client_add_cmd_option(client, opt);
 163
 164         /*
 165          * Duplicate arguments to shared client group
 166          */
 167         flist_for_each(entry, &arg_list) {
 168                 client = flist_entry(entry, struct fio_client, arg_list);
 169
 170                 __fio_client_add_cmd_option(client, opt);
 171         }
 172 }
 173
 174 int fio_client_add(const char *hostname, void **cookie)
 175 {
 176         struct fio_client *existing = *cookie;
 177         struct fio_client *client;
 178
 179         if (existing) {
 180                 /*
 181                  * We always add our "exec" name as the option, hence 1
 182                  * means empty.
 183                  */
 184                 if (existing->argc == 1)
 185                         flist_add_tail(&existing->arg_list, &arg_list);
 186                 else {
 187                         while (!flist_empty(&arg_list))
 188                                 flist_del_init(arg_list.next);
 189                 }
 190         }
 191
 192         client = malloc(sizeof(*client));
 193         memset(client, 0, sizeof(*client));
 194
 195         INIT_FLIST_HEAD(&client->list);
 196         INIT_FLIST_HEAD(&client->hash_list);
 197         INIT_FLIST_HEAD(&client->arg_list);
 198         INIT_FLIST_HEAD(&client->eta_list);
 199         INIT_FLIST_HEAD(&client->cmd_list);
 200
 201         if (fio_server_parse_string(hostname, &client->hostname,
 202                                         &client->is_sock, &client->port,
 203                                         &client->addr.sin_addr))
 204                 return -1;
 205
 206         client->fd = -1;
 207
 208         __fio_client_add_cmd_option(client, "fio");
 209
 210         flist_add(&client->list, &client_list);
 211         nr_clients++;
 212         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 213         *cookie = client;
 214         return 0;
 215 }
 216
 217 static int fio_client_connect_ip(struct fio_client *client)
 218 {
 219         int fd;
 220
 221         client->addr.sin_family = AF_INET;
 222         client->addr.sin_port = htons(client->port);
 223
 224         fd = socket(AF_INET, SOCK_STREAM, 0);
 225         if (fd < 0) {
 226                 log_err("fio: socket: %s\n", strerror(errno));
 227                 return -1;
 228         }
 229
 230         if (connect(fd, (struct sockaddr *) &client->addr, sizeof(client->addr)) < 0) {
 231                 log_err("fio: connect: %s\n", strerror(errno));
 232                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 233                                                                 client->port);
 234                 close(fd);
 235                 return -1;
 236         }
 237
 238         return fd;
 239 }
 240
 241 static int fio_client_connect_sock(struct fio_client *client)
 242 {
 243         struct sockaddr_un *addr = &client->addr_un;
 244         fio_socklen_t len;
 245         int fd;
 246
 247         memset(addr, 0, sizeof(*addr));
 248         addr->sun_family = AF_UNIX;
 249         strcpy(addr->sun_path, client->hostname);
 250
 251         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 252         if (fd < 0) {
 253                 log_err("fio: socket: %s\n", strerror(errno));
 254                 return -1;
 255         }
 256
 257         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 258         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 259                 log_err("fio: connect; %s\n", strerror(errno));
 260                 close(fd);
 261                 return -1;
 262         }
 263
 264         return fd;
 265 }
 266
 267 static int fio_client_connect(struct fio_client *client)
 268 {
 269         int fd;
 270
 271         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 272
 273         if (client->is_sock)
 274                 fd = fio_client_connect_sock(client);
 275         else
 276                 fd = fio_client_connect_ip(client);
 277
 278         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 279
 280         if (fd < 0)
 281                 return 1;
 282
 283         client->fd = fd;
 284         fio_client_add_hash(client);
 285         client->state = Client_connected;
 286         return 0;
 287 }
 288
 289 void fio_clients_terminate(void)
 290 {
 291         struct flist_head *entry;
 292         struct fio_client *client;
 293
 294         dprint(FD_NET, "client: terminate clients\n");
 295
 296         flist_for_each(entry, &client_list) {
 297                 client = flist_entry(entry, struct fio_client, list);
 298
 299                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_QUIT, 0, NULL);
 300         }
 301 }
 302
 303 static void sig_int(int sig)
 304 {
 305         dprint(FD_NET, "client: got signal %d\n", sig);
 306         fio_clients_terminate();
 307 }
 308
 309 static void client_signal_handler(void)
 310 {
 311         struct sigaction act;
 312
 313         memset(&act, 0, sizeof(act));
 314         act.sa_handler = sig_int;
 315         act.sa_flags = SA_RESTART;
 316         sigaction(SIGINT, &act, NULL);
 317
 318         memset(&act, 0, sizeof(act));
 319         act.sa_handler = sig_int;
 320         act.sa_flags = SA_RESTART;
 321         sigaction(SIGTERM, &act, NULL);
 322 }
 323
 324 static void probe_client(struct fio_client *client)
 325 {
 326         dprint(FD_NET, "client: send probe\n");
 327
 328         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
 329 }
 330
 331 static int send_client_cmd_line(struct fio_client *client)
 332 {
 333         struct cmd_single_line_pdu *cslp;
 334         struct cmd_line_pdu *clp;
 335         unsigned long offset;
 336         unsigned int *lens;
 337         void *pdu;
 338         size_t mem;
 339         int i, ret;
 340
 341         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 342
 343         lens = malloc(client->argc * sizeof(unsigned int));
 344
 345         /*
 346          * Find out how much mem we need
 347          */
 348         for (i = 0, mem = 0; i < client->argc; i++) {
 349                 lens[i] = strlen(client->argv[i]) + 1;
 350                 mem += lens[i];
 351         }
 352
 353         /*
 354          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 355          */
 356         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 357
 358         pdu = malloc(mem);
 359         clp = pdu;
 360         offset = sizeof(*clp);
 361
 362         for (i = 0; i < client->argc; i++) {
 363                 uint16_t arg_len = lens[i];
 364
 365                 cslp = pdu + offset;
 366                 strcpy((char *) cslp->text, client->argv[i]);
 367                 cslp->len = cpu_to_le16(arg_len);
 368                 offset += sizeof(*cslp) + arg_len;
 369         }
 370
 371         free(lens);
 372         clp->lines = cpu_to_le16(client->argc);
 373         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
 374         free(pdu);
 375         return ret;
 376 }
 377
 378 int fio_clients_connect(void)
 379 {
 380         struct fio_client *client;
 381         struct flist_head *entry, *tmp;
 382         int ret;
 383
 384         dprint(FD_NET, "client: connect all\n");
 385
 386         client_signal_handler();
 387
 388         flist_for_each_safe(entry, tmp, &client_list) {
 389                 client = flist_entry(entry, struct fio_client, list);
 390
 391                 ret = fio_client_connect(client);
 392                 if (ret) {
 393                         remove_client(client);
 394                         continue;
 395                 }
 396
 397                 probe_client(client);
 398
 399                 if (client->argc > 1)
 400                         send_client_cmd_line(client);
 401         }
 402
 403         return !nr_clients;
 404 }
 405
 406 /*
 407  * Send file contents to server backend. We could use sendfile(), but to remain
 408  * more portable lets just read/write the darn thing.
 409  */
 410 static int fio_client_send_ini(struct fio_client *client, const char *filename)
 411 {
 412         struct stat sb;
 413         char *p, *buf;
 414         off_t len;
 415         int fd, ret;
 416
 417         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 418
 419         fd = open(filename, O_RDONLY);
 420         if (fd < 0) {
 421                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 422                 return 1;
 423         }
 424
 425         if (fstat(fd, &sb) < 0) {
 426                 log_err("fio: job file stat: %s\n", strerror(errno));
 427                 close(fd);
 428                 return 1;
 429         }
 430
 431         buf = malloc(sb.st_size);
 432
 433         len = sb.st_size;
 434         p = buf;
 435         do {
 436                 ret = read(fd, p, len);
 437                 if (ret > 0) {
 438                         len -= ret;
 439                         if (!len)
 440                                 break;
 441                         p += ret;
 442                         continue;
 443                 } else if (!ret)
 444                         break;
 445                 else if (errno == EAGAIN || errno == EINTR)
 446                         continue;
 447         } while (1);
 448
 449         if (len) {
 450                 log_err("fio: failed reading job file %s\n", filename);
 451                 close(fd);
 452                 free(buf);
 453                 return 1;
 454         }
 455
 456         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, buf, sb.st_size, 0);
 457         free(buf);
 458         close(fd);
 459         return ret;
 460 }
 461
 462 int fio_clients_send_ini(const char *filename)
 463 {
 464         struct fio_client *client;
 465         struct flist_head *entry, *tmp;
 466
 467         flist_for_each_safe(entry, tmp, &client_list) {
 468                 client = flist_entry(entry, struct fio_client, list);
 469
 470                 if (fio_client_send_ini(client, filename))
 471                         remove_client(client);
 472         }
 473
 474         return !nr_clients;
 475 }
 476
 477 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 478 {
 479         dst->max_val    = le64_to_cpu(src->max_val);
 480         dst->min_val    = le64_to_cpu(src->min_val);
 481         dst->samples    = le64_to_cpu(src->samples);
 482
 483         /*
 484          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 485          */
 486         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 487         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 488 }
 489
 490 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 491 {
 492         int i, j;
 493
 494         dst->error      = le32_to_cpu(src->error);
 495         dst->groupid    = le32_to_cpu(src->groupid);
 496         dst->pid        = le32_to_cpu(src->pid);
 497         dst->members    = le32_to_cpu(src->members);
 498
 499         for (i = 0; i < 2; i++) {
 500                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 501                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 502                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 503                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 504         }
 505
 506         dst->usr_time           = le64_to_cpu(src->usr_time);
 507         dst->sys_time           = le64_to_cpu(src->sys_time);
 508         dst->ctx                = le64_to_cpu(src->ctx);
 509         dst->minf               = le64_to_cpu(src->minf);
 510         dst->majf               = le64_to_cpu(src->majf);
 511         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 512
 513         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 514                 fio_fp64_t *fps = &src->percentile_list[i];
 515                 fio_fp64_t *fpd = &dst->percentile_list[i];
 516
 517                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 518         }
 519
 520         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 521                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 522                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 523                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 524         }
 525
 526         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 527                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 528                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 529         }
 530
 531         for (i = 0; i < 2; i++)
 532                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 533                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 534
 535         for (i = 0; i < 3; i++) {
 536                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 537                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 538         }
 539
 540         dst->total_submit       = le64_to_cpu(src->total_submit);
 541         dst->total_complete     = le64_to_cpu(src->total_complete);
 542
 543         for (i = 0; i < 2; i++) {
 544                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 545                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 546         }
 547
 548         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 549         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 550         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 551         dst->first_error        = le32_to_cpu(src->first_error);
 552         dst->kb_base            = le32_to_cpu(src->kb_base);
 553 }
 554
 555 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 556 {
 557         int i;
 558
 559         for (i = 0; i < 2; i++) {
 560                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 561                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 562                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 563                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 564                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 565                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 566         }
 567
 568         dst->kb_base    = le32_to_cpu(src->kb_base);
 569         dst->groupid    = le32_to_cpu(src->groupid);
 570 }
 571
 572 static void handle_ts(struct fio_net_cmd *cmd)
 573 {
 574         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 575
 576         convert_ts(&p->ts, &p->ts);
 577         convert_gs(&p->rs, &p->rs);
 578
 579         show_thread_status(&p->ts, &p->rs);
 580
 581         if (sum_stat_clients == 1)
 582                 return;
 583
 584         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 585         sum_group_stats(&client_gs, &p->rs);
 586
 587         client_ts.members++;
 588         client_ts.groupid = p->ts.groupid;
 589
 590         if (++sum_stat_nr == sum_stat_clients) {
 591                 strcpy(client_ts.name, "All clients");
 592                 show_thread_status(&client_ts, &client_gs);
 593         }
 594 }
 595
 596 static void handle_gs(struct fio_net_cmd *cmd)
 597 {
 598         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 599
 600         convert_gs(gs, gs);
 601         show_group_stats(gs);
 602 }
 603
 604 static void convert_jobs_eta(struct jobs_eta *je)
 605 {
 606         int i;
 607
 608         je->nr_running          = le32_to_cpu(je->nr_running);
 609         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 610         je->nr_pending          = le32_to_cpu(je->nr_pending);
 611         je->files_open          = le32_to_cpu(je->files_open);
 612         je->m_rate              = le32_to_cpu(je->m_rate);
 613         je->t_rate              = le32_to_cpu(je->t_rate);
 614         je->m_iops              = le32_to_cpu(je->m_iops);
 615         je->t_iops              = le32_to_cpu(je->t_iops);
 616
 617         for (i = 0; i < 2; i++) {
 618                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 619                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 620         }
 621
 622         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 623         je->eta_sec             = le64_to_cpu(je->eta_sec);
 624 }
 625
 626 static void sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 627 {
 628         int i;
 629
 630         dst->nr_running         += je->nr_running;
 631         dst->nr_ramp            += je->nr_ramp;
 632         dst->nr_pending         += je->nr_pending;
 633         dst->files_open         += je->files_open;
 634         dst->m_rate             += je->m_rate;
 635         dst->t_rate             += je->t_rate;
 636         dst->m_iops             += je->m_iops;
 637         dst->t_iops             += je->t_iops;
 638
 639         for (i = 0; i < 2; i++) {
 640                 dst->rate[i]    += je->rate[i];
 641                 dst->iops[i]    += je->iops[i];
 642         }
 643
 644         dst->elapsed_sec        += je->elapsed_sec;
 645
 646         if (je->eta_sec > dst->eta_sec)
 647                 dst->eta_sec = je->eta_sec;
 648 }
 649
 650 static void dec_jobs_eta(struct client_eta *eta)
 651 {
 652         if (!--eta->pending) {
 653                 display_thread_status(&eta->eta);
 654                 free(eta);
 655         }
 656 }
 657
 658 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 659 {
 660         struct fio_net_int_cmd *icmd = NULL;
 661         struct flist_head *entry;
 662
 663         flist_for_each(entry, &client->cmd_list) {
 664                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
 665
 666                 if (cmd->tag == (uintptr_t) icmd)
 667                         break;
 668
 669                 icmd = NULL;
 670         }
 671
 672         if (!icmd) {
 673                 log_err("fio: client: unable to find matching tag\n");
 674                 return;
 675         }
 676
 677         flist_del(&icmd->list);
 678         cmd->tag = icmd->saved_tag;
 679         free(icmd);
 680 }
 681
 682 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 683 {
 684         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 685         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 686
 687         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 688
 689         assert(client->eta_in_flight == eta);
 690
 691         client->eta_in_flight = NULL;
 692         flist_del_init(&client->eta_list);
 693
 694         convert_jobs_eta(je);
 695         sum_jobs_eta(&eta->eta, je);
 696         dec_jobs_eta(eta);
 697 }
 698
 699 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 700 {
 701         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 702         const char *os, *arch;
 703         char bit[16];
 704
 705         os = fio_get_os_string(probe->os);
 706         if (!os)
 707                 os = "unknown";
 708
 709         arch = fio_get_arch_string(probe->arch);
 710         if (!arch)
 711                 os = "unknown";
 712
 713         sprintf(bit, "%d-bit", probe->bpp * 8);
 714
 715         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
 716                 probe->hostname, probe->bigendian, bit, os, arch,
 717                 probe->fio_major, probe->fio_minor, probe->fio_patch);
 718
 719         if (!client->name)
 720                 client->name = strdup((char *) probe->hostname);
 721 }
 722
 723 static int handle_client(struct fio_client *client)
 724 {
 725         struct fio_net_cmd *cmd;
 726
 727         dprint(FD_NET, "client: handle %s\n", client->hostname);
 728
 729         cmd = fio_net_recv_cmd(client->fd);
 730         if (!cmd)
 731                 return 0;
 732
 733         dprint(FD_NET, "client: got cmd op %s from %s\n",
 734                                 fio_server_op(cmd->opcode), client->hostname);
 735
 736         switch (cmd->opcode) {
 737         case FIO_NET_CMD_QUIT:
 738                 remove_client(client);
 739                 free(cmd);
 740                 break;
 741         case FIO_NET_CMD_TEXT: {
 742                 const char *buf = (const char *) cmd->payload;
 743                 const char *name;
 744                 int fio_unused ret;
 745
 746                 name = client->name ? client->name : client->hostname;
 747
 748                 if (!client->skip_newline)
 749                         fprintf(f_out, "<%s> ", name);
 750                 ret = fwrite(buf, cmd->pdu_len, 1, f_out);
 751                 fflush(f_out);
 752                 client->skip_newline = strchr(buf, '\n') == NULL;
 753                 free(cmd);
 754                 break;
 755                 }
 756         case FIO_NET_CMD_TS:
 757                 handle_ts(cmd);
 758                 free(cmd);
 759                 break;
 760         case FIO_NET_CMD_GS:
 761                 handle_gs(cmd);
 762                 free(cmd);
 763                 break;
 764         case FIO_NET_CMD_ETA:
 765                 remove_reply_cmd(client, cmd);
 766                 handle_eta(client, cmd);
 767                 free(cmd);
 768                 break;
 769         case FIO_NET_CMD_PROBE:
 770                 remove_reply_cmd(client, cmd);
 771                 handle_probe(client, cmd);
 772                 free(cmd);
 773                 break;
 774         case FIO_NET_CMD_START:
 775                 client->state = Client_started;
 776                 free(cmd);
 777                 break;
 778         case FIO_NET_CMD_STOP:
 779                 client->state = Client_stopped;
 780                 free(cmd);
 781                 break;
 782         default:
 783                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
 784                 free(cmd);
 785                 break;
 786         }
 787
 788         return 1;
 789 }
 790
 791 static void request_client_etas(void)
 792 {
 793         struct fio_client *client;
 794         struct flist_head *entry;
 795         struct client_eta *eta;
 796         int skipped = 0;
 797
 798         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
 799
 800         eta = malloc(sizeof(*eta));
 801         memset(&eta->eta, 0, sizeof(eta->eta));
 802         eta->pending = nr_clients;
 803
 804         flist_for_each(entry, &client_list) {
 805                 client = flist_entry(entry, struct fio_client, list);
 806
 807                 if (!flist_empty(&client->eta_list)) {
 808                         skipped++;
 809                         continue;
 810                 }
 811
 812                 assert(!client->eta_in_flight);
 813                 flist_add_tail(&client->eta_list, &eta_list);
 814                 client->eta_in_flight = eta;
 815                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
 816                                         (uintptr_t) eta, &client->cmd_list);
 817         }
 818
 819         while (skipped--)
 820                 dec_jobs_eta(eta);
 821
 822         dprint(FD_NET, "client: requested eta tag %p\n", eta);
 823 }
 824
 825 static int client_check_cmd_timeout(struct fio_client *client,
 826                                     struct timeval *now)
 827 {
 828         struct fio_net_int_cmd *cmd;
 829         struct flist_head *entry, *tmp;
 830         int ret = 0;
 831
 832         flist_for_each_safe(entry, tmp, &client->cmd_list) {
 833                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
 834
 835                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
 836                         continue;
 837
 838                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
 839                                                 fio_server_op(cmd->cmd.opcode));
 840                 flist_del(&cmd->list);
 841                 free(cmd);
 842                 ret = 1;
 843         }
 844
 845         return flist_empty(&client->cmd_list) && ret;
 846 }
 847
 848 static int fio_client_timed_out(void)
 849 {
 850         struct fio_client *client;
 851         struct flist_head *entry, *tmp;
 852         struct timeval tv;
 853         int ret = 0;
 854
 855         gettimeofday(&tv, NULL);
 856
 857         flist_for_each_safe(entry, tmp, &client_list) {
 858                 client = flist_entry(entry, struct fio_client, list);
 859
 860                 if (flist_empty(&client->cmd_list))
 861                         continue;
 862
 863                 if (!client_check_cmd_timeout(client, &tv))
 864                         continue;
 865
 866                 log_err("fio: client %s timed out\n", client->hostname);
 867                 remove_client(client);
 868                 ret = 1;
 869         }
 870
 871         return ret;
 872 }
 873
 874 int fio_handle_clients(void)
 875 {
 876         struct fio_client *client;
 877         struct flist_head *entry;
 878         struct pollfd *pfds;
 879         int i, ret = 0;
 880
 881         gettimeofday(&eta_tv, NULL);
 882
 883         pfds = malloc(nr_clients * sizeof(struct pollfd));
 884
 885         sum_stat_clients = nr_clients;
 886         init_thread_stat(&client_ts);
 887         init_group_run_stat(&client_gs);
 888
 889         while (!exit_backend && nr_clients) {
 890                 i = 0;
 891                 flist_for_each(entry, &client_list) {
 892                         client = flist_entry(entry, struct fio_client, list);
 893
 894                         pfds[i].fd = client->fd;
 895                         pfds[i].events = POLLIN;
 896                         i++;
 897                 }
 898
 899                 assert(i == nr_clients);
 900
 901                 do {
 902                         struct timeval tv;
 903
 904                         gettimeofday(&tv, NULL);
 905                         if (mtime_since(&eta_tv, &tv) >= 900) {
 906                                 request_client_etas();
 907                                 memcpy(&eta_tv, &tv, sizeof(tv));
 908
 909                                 if (fio_client_timed_out())
 910                                         break;
 911                         }
 912
 913                         ret = poll(pfds, nr_clients, 100);
 914                         if (ret < 0) {
 915                                 if (errno == EINTR)
 916                                         continue;
 917                                 log_err("fio: poll clients: %s\n", strerror(errno));
 918                                 break;
 919                         } else if (!ret)
 920                                 continue;
 921                 } while (ret <= 0);
 922
 923                 for (i = 0; i < nr_clients; i++) {
 924                         if (!(pfds[i].revents & POLLIN))
 925                                 continue;
 926
 927                         client = find_client_by_fd(pfds[i].fd);
 928                         if (!client) {
 929                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
 930                                 continue;
 931                         }
 932                         if (!handle_client(client)) {
 933                                 log_info("client: host=%s disconnected\n",
 934                                                 client->hostname);
 935                                 remove_client(client);
 936                         }
 937                 }
 938         }
 939
 940         free(pfds);
 941         return 0;
 942 }