client.c

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