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 #include <zlib.h>
  18
  19 #include "fio.h"
  20 #include "client.h"
  21 #include "server.h"
  22 #include "flist.h"
  23 #include "hash.h"
  24
  25 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd);
  26 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd);
  27 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd);
  28 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd);
  29 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd);
  30 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd);
  31 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd);
  32
  33 struct client_ops fio_client_ops = {
  34         .text           = handle_text,
  35         .disk_util      = handle_du,
  36         .thread_status  = handle_ts,
  37         .group_stats    = handle_gs,
  38         .stop           = handle_stop,
  39         .start          = handle_start,
  40         .eta            = display_thread_status,
  41         .probe          = handle_probe,
  42         .eta_msec       = FIO_CLIENT_DEF_ETA_MSEC,
  43         .client_type    = FIO_CLIENT_TYPE_CLI,
  44 };
  45
  46 static struct timeval eta_tv;
  47
  48 static FLIST_HEAD(client_list);
  49 static FLIST_HEAD(eta_list);
  50
  51 static FLIST_HEAD(arg_list);
  52
  53 struct thread_stat client_ts;
  54 struct group_run_stats client_gs;
  55 int sum_stat_clients;
  56
  57 static int sum_stat_nr;
  58
  59 #define FIO_CLIENT_HASH_BITS    7
  60 #define FIO_CLIENT_HASH_SZ      (1 << FIO_CLIENT_HASH_BITS)
  61 #define FIO_CLIENT_HASH_MASK    (FIO_CLIENT_HASH_SZ - 1)
  62 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ];
  63
  64 static void fio_client_add_hash(struct fio_client *client)
  65 {
  66         int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS);
  67
  68         bucket &= FIO_CLIENT_HASH_MASK;
  69         flist_add(&client->hash_list, &client_hash[bucket]);
  70 }
  71
  72 static void fio_client_remove_hash(struct fio_client *client)
  73 {
  74         if (!flist_empty(&client->hash_list))
  75                 flist_del_init(&client->hash_list);
  76 }
  77
  78 static void fio_init fio_client_hash_init(void)
  79 {
  80         int i;
  81
  82         for (i = 0; i < FIO_CLIENT_HASH_SZ; i++)
  83                 INIT_FLIST_HEAD(&client_hash[i]);
  84 }
  85
  86 static struct fio_client *find_client_by_fd(int fd)
  87 {
  88         int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK;
  89         struct fio_client *client;
  90         struct flist_head *entry;
  91
  92         flist_for_each(entry, &client_hash[bucket]) {
  93                 client = flist_entry(entry, struct fio_client, hash_list);
  94
  95                 if (client->fd == fd) {
  96                         client->refs++;
  97                         return client;
  98                 }
  99         }
 100
 101         return NULL;
 102 }
 103
 104 void fio_put_client(struct fio_client *client)
 105 {
 106         if (--client->refs)
 107                 return;
 108
 109         free(client->hostname);
 110         if (client->argv)
 111                 free(client->argv);
 112         if (client->name)
 113                 free(client->name);
 114
 115         free(client);
 116 }
 117
 118 static void remove_client(struct fio_client *client)
 119 {
 120         assert(client->refs);
 121
 122         dprint(FD_NET, "client: removed <%s>\n", client->hostname);
 123
 124         if (!flist_empty(&client->list))
 125                 flist_del_init(&client->list);
 126
 127         fio_client_remove_hash(client);
 128
 129         if (!flist_empty(&client->eta_list)) {
 130                 flist_del_init(&client->eta_list);
 131                 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta);
 132         }
 133
 134         close(client->fd);
 135         client->fd = -1;
 136
 137         if (client->ops->removed)
 138                 client->ops->removed(client);
 139
 140         nr_clients--;
 141         sum_stat_clients--;
 142
 143         fio_put_client(client);
 144 }
 145
 146 struct fio_client *fio_get_client(struct fio_client *client)
 147 {
 148         client->refs++;
 149         return client;
 150 }
 151
 152 static void __fio_client_add_cmd_option(struct fio_client *client,
 153                                         const char *opt)
 154 {
 155         int index;
 156
 157         index = client->argc++;
 158         client->argv = realloc(client->argv, sizeof(char *) * client->argc);
 159         client->argv[index] = strdup(opt);
 160         dprint(FD_NET, "client: add cmd %d: %s\n", index, opt);
 161 }
 162
 163 void fio_client_add_cmd_option(void *cookie, const char *opt)
 164 {
 165         struct fio_client *client = cookie;
 166         struct flist_head *entry;
 167
 168         if (!client || !opt)
 169                 return;
 170
 171         __fio_client_add_cmd_option(client, opt);
 172
 173         /*
 174          * Duplicate arguments to shared client group
 175          */
 176         flist_for_each(entry, &arg_list) {
 177                 client = flist_entry(entry, struct fio_client, arg_list);
 178
 179                 __fio_client_add_cmd_option(client, opt);
 180         }
 181 }
 182
 183 struct fio_client *fio_client_add_explicit(struct client_ops *ops,
 184                                            const char *hostname, int type,
 185                                            int port)
 186 {
 187         struct fio_client *client;
 188
 189         client = malloc(sizeof(*client));
 190         memset(client, 0, sizeof(*client));
 191
 192         INIT_FLIST_HEAD(&client->list);
 193         INIT_FLIST_HEAD(&client->hash_list);
 194         INIT_FLIST_HEAD(&client->arg_list);
 195         INIT_FLIST_HEAD(&client->eta_list);
 196         INIT_FLIST_HEAD(&client->cmd_list);
 197
 198         client->hostname = strdup(hostname);
 199
 200         if (type == Fio_client_socket)
 201                 client->is_sock = 1;
 202         else {
 203                 int ipv6;
 204
 205                 ipv6 = type == Fio_client_ipv6;
 206                 if (fio_server_parse_host(hostname, &ipv6,
 207                                                 &client->addr.sin_addr,
 208                                                 &client->addr6.sin6_addr))
 209                         goto err;
 210
 211                 client->port = port;
 212         }
 213
 214         client->fd = -1;
 215         client->ops = ops;
 216         client->refs = 1;
 217         client->type = ops->client_type;
 218
 219         __fio_client_add_cmd_option(client, "fio");
 220
 221         flist_add(&client->list, &client_list);
 222         nr_clients++;
 223         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 224         return client;
 225 err:
 226         free(client);
 227         return NULL;
 228 }
 229
 230 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie)
 231 {
 232         struct fio_client *existing = *cookie;
 233         struct fio_client *client;
 234
 235         if (existing) {
 236                 /*
 237                  * We always add our "exec" name as the option, hence 1
 238                  * means empty.
 239                  */
 240                 if (existing->argc == 1)
 241                         flist_add_tail(&existing->arg_list, &arg_list);
 242                 else {
 243                         while (!flist_empty(&arg_list))
 244                                 flist_del_init(arg_list.next);
 245                 }
 246         }
 247
 248         client = malloc(sizeof(*client));
 249         memset(client, 0, sizeof(*client));
 250
 251         INIT_FLIST_HEAD(&client->list);
 252         INIT_FLIST_HEAD(&client->hash_list);
 253         INIT_FLIST_HEAD(&client->arg_list);
 254         INIT_FLIST_HEAD(&client->eta_list);
 255         INIT_FLIST_HEAD(&client->cmd_list);
 256
 257         if (fio_server_parse_string(hostname, &client->hostname,
 258                                         &client->is_sock, &client->port,
 259                                         &client->addr.sin_addr,
 260                                         &client->addr6.sin6_addr,
 261                                         &client->ipv6))
 262                 return -1;
 263
 264         client->fd = -1;
 265         client->ops = ops;
 266         client->refs = 1;
 267         client->type = ops->client_type;
 268
 269         __fio_client_add_cmd_option(client, "fio");
 270
 271         flist_add(&client->list, &client_list);
 272         nr_clients++;
 273         dprint(FD_NET, "client: added <%s>\n", client->hostname);
 274         *cookie = client;
 275         return 0;
 276 }
 277
 278 static void probe_client(struct fio_client *client)
 279 {
 280         dprint(FD_NET, "client: send probe\n");
 281
 282         fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_PROBE, 0, &client->cmd_list);
 283 }
 284
 285 static int fio_client_connect_ip(struct fio_client *client)
 286 {
 287         struct sockaddr *addr;
 288         fio_socklen_t socklen;
 289         int fd, domain;
 290
 291         if (client->ipv6) {
 292                 client->addr6.sin6_family = AF_INET6;
 293                 client->addr6.sin6_port = htons(client->port);
 294                 domain = AF_INET6;
 295                 addr = (struct sockaddr *) &client->addr6;
 296                 socklen = sizeof(client->addr6);
 297         } else {
 298                 client->addr.sin_family = AF_INET;
 299                 client->addr.sin_port = htons(client->port);
 300                 domain = AF_INET;
 301                 addr = (struct sockaddr *) &client->addr;
 302                 socklen = sizeof(client->addr);
 303         }
 304
 305         fd = socket(domain, SOCK_STREAM, 0);
 306         if (fd < 0) {
 307                 int ret = -errno;
 308
 309                 log_err("fio: socket: %s\n", strerror(errno));
 310                 return ret;
 311         }
 312
 313         if (connect(fd, addr, socklen) < 0) {
 314                 int ret = -errno;
 315
 316                 log_err("fio: connect: %s\n", strerror(errno));
 317                 log_err("fio: failed to connect to %s:%u\n", client->hostname,
 318                                                                 client->port);
 319                 close(fd);
 320                 return ret;
 321         }
 322
 323         return fd;
 324 }
 325
 326 static int fio_client_connect_sock(struct fio_client *client)
 327 {
 328         struct sockaddr_un *addr = &client->addr_un;
 329         fio_socklen_t len;
 330         int fd;
 331
 332         memset(addr, 0, sizeof(*addr));
 333         addr->sun_family = AF_UNIX;
 334         strcpy(addr->sun_path, client->hostname);
 335
 336         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 337         if (fd < 0) {
 338                 int ret = -errno;
 339
 340                 log_err("fio: socket: %s\n", strerror(errno));
 341                 return ret;
 342         }
 343
 344         len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1;
 345         if (connect(fd, (struct sockaddr *) addr, len) < 0) {
 346                 int ret = -errno;
 347
 348                 log_err("fio: connect; %s\n", strerror(errno));
 349                 close(fd);
 350                 return ret;
 351         }
 352
 353         return fd;
 354 }
 355
 356 int fio_client_connect(struct fio_client *client)
 357 {
 358         int fd;
 359
 360         dprint(FD_NET, "client: connect to host %s\n", client->hostname);
 361
 362         if (client->is_sock)
 363                 fd = fio_client_connect_sock(client);
 364         else
 365                 fd = fio_client_connect_ip(client);
 366
 367         dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd);
 368
 369         if (fd < 0)
 370                 return fd;
 371
 372         client->fd = fd;
 373         fio_client_add_hash(client);
 374         client->state = Client_connected;
 375
 376         probe_client(client);
 377         return 0;
 378 }
 379
 380 int fio_client_terminate(struct fio_client *client)
 381 {
 382         return fio_net_send_quit(client->fd);
 383 }
 384
 385 void fio_clients_terminate(void)
 386 {
 387         struct flist_head *entry;
 388         struct fio_client *client;
 389
 390         dprint(FD_NET, "client: terminate clients\n");
 391
 392         flist_for_each(entry, &client_list) {
 393                 client = flist_entry(entry, struct fio_client, list);
 394                 fio_client_terminate(client);
 395         }
 396 }
 397
 398 static void sig_int(int sig)
 399 {
 400         dprint(FD_NET, "client: got signal %d\n", sig);
 401         fio_clients_terminate();
 402 }
 403
 404 static void client_signal_handler(void)
 405 {
 406         struct sigaction act;
 407
 408         memset(&act, 0, sizeof(act));
 409         act.sa_handler = sig_int;
 410         act.sa_flags = SA_RESTART;
 411         sigaction(SIGINT, &act, NULL);
 412
 413         memset(&act, 0, sizeof(act));
 414         act.sa_handler = sig_int;
 415         act.sa_flags = SA_RESTART;
 416         sigaction(SIGTERM, &act, NULL);
 417 }
 418
 419 static int send_client_cmd_line(struct fio_client *client)
 420 {
 421         struct cmd_single_line_pdu *cslp;
 422         struct cmd_line_pdu *clp;
 423         unsigned long offset;
 424         unsigned int *lens;
 425         void *pdu;
 426         size_t mem;
 427         int i, ret;
 428
 429         dprint(FD_NET, "client: send cmdline %d\n", client->argc);
 430
 431         lens = malloc(client->argc * sizeof(unsigned int));
 432
 433         /*
 434          * Find out how much mem we need
 435          */
 436         for (i = 0, mem = 0; i < client->argc; i++) {
 437                 lens[i] = strlen(client->argv[i]) + 1;
 438                 mem += lens[i];
 439         }
 440
 441         /*
 442          * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu
 443          */
 444         mem += sizeof(*clp) + (client->argc * sizeof(*cslp));
 445
 446         pdu = malloc(mem);
 447         clp = pdu;
 448         offset = sizeof(*clp);
 449
 450         for (i = 0; i < client->argc; i++) {
 451                 uint16_t arg_len = lens[i];
 452
 453                 cslp = pdu + offset;
 454                 strcpy((char *) cslp->text, client->argv[i]);
 455                 cslp->len = cpu_to_le16(arg_len);
 456                 offset += sizeof(*cslp) + arg_len;
 457         }
 458
 459         free(lens);
 460         clp->lines = cpu_to_le16(client->argc);
 461         clp->client_type = __cpu_to_le16(client->type);
 462         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, 0);
 463         free(pdu);
 464         return ret;
 465 }
 466
 467 int fio_clients_connect(void)
 468 {
 469         struct fio_client *client;
 470         struct flist_head *entry, *tmp;
 471         int ret;
 472
 473 #ifdef WIN32
 474         WSADATA wsd;
 475         WSAStartup(MAKEWORD(2,2), &wsd);
 476 #endif
 477
 478         dprint(FD_NET, "client: connect all\n");
 479
 480         client_signal_handler();
 481
 482         flist_for_each_safe(entry, tmp, &client_list) {
 483                 client = flist_entry(entry, struct fio_client, list);
 484
 485                 ret = fio_client_connect(client);
 486                 if (ret) {
 487                         remove_client(client);
 488                         continue;
 489                 }
 490
 491                 if (client->argc > 1)
 492                         send_client_cmd_line(client);
 493         }
 494
 495         return !nr_clients;
 496 }
 497
 498 int fio_start_client(struct fio_client *client)
 499 {
 500         dprint(FD_NET, "client: start %s\n", client->hostname);
 501         return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL);
 502 }
 503
 504 int fio_start_all_clients(void)
 505 {
 506         struct fio_client *client;
 507         struct flist_head *entry, *tmp;
 508         int ret;
 509
 510         dprint(FD_NET, "client: start all\n");
 511
 512         flist_for_each_safe(entry, tmp, &client_list) {
 513                 client = flist_entry(entry, struct fio_client, list);
 514
 515                 ret = fio_start_client(client);
 516                 if (ret) {
 517                         remove_client(client);
 518                         continue;
 519                 }
 520         }
 521
 522         return flist_empty(&client_list);
 523 }
 524
 525 /*
 526  * Send file contents to server backend. We could use sendfile(), but to remain
 527  * more portable lets just read/write the darn thing.
 528  */
 529 static int __fio_client_send_ini(struct fio_client *client, const char *filename)
 530 {
 531         struct cmd_job_pdu *pdu;
 532         size_t p_size;
 533         struct stat sb;
 534         char *p;
 535         void *buf;
 536         off_t len;
 537         int fd, ret;
 538
 539         dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname);
 540
 541         fd = open(filename, O_RDONLY);
 542         if (fd < 0) {
 543                 int ret = -errno;
 544
 545                 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno));
 546                 return ret;
 547         }
 548
 549         if (fstat(fd, &sb) < 0) {
 550                 int ret = -errno;
 551
 552                 log_err("fio: job file stat: %s\n", strerror(errno));
 553                 close(fd);
 554                 return ret;
 555         }
 556
 557         p_size = sb.st_size + sizeof(*pdu);
 558         pdu = malloc(p_size);
 559         buf = pdu->buf;
 560
 561         len = sb.st_size;
 562         p = buf;
 563         do {
 564                 ret = read(fd, p, len);
 565                 if (ret > 0) {
 566                         len -= ret;
 567                         if (!len)
 568                                 break;
 569                         p += ret;
 570                         continue;
 571                 } else if (!ret)
 572                         break;
 573                 else if (errno == EAGAIN || errno == EINTR)
 574                         continue;
 575         } while (1);
 576
 577         if (len) {
 578                 log_err("fio: failed reading job file %s\n", filename);
 579                 close(fd);
 580                 free(buf);
 581                 return 1;
 582         }
 583
 584         pdu->buf_len = __cpu_to_le32(sb.st_size);
 585         pdu->client_type = cpu_to_le32(client->type);
 586
 587         client->sent_job = 1;
 588         ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, 0);
 589         free(pdu);
 590         close(fd);
 591         return ret;
 592 }
 593
 594 int fio_client_send_ini(struct fio_client *client, const char *filename)
 595 {
 596         int ret;
 597
 598         ret = __fio_client_send_ini(client, filename);
 599         if (!ret)
 600                 client->sent_job = 1;
 601
 602         return ret;
 603 }
 604
 605 int fio_clients_send_ini(const char *filename)
 606 {
 607         struct fio_client *client;
 608         struct flist_head *entry, *tmp;
 609
 610         flist_for_each_safe(entry, tmp, &client_list) {
 611                 client = flist_entry(entry, struct fio_client, list);
 612
 613                 if (fio_client_send_ini(client, filename))
 614                         remove_client(client);
 615         }
 616
 617         return !nr_clients;
 618 }
 619
 620 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 621 {
 622         dst->max_val    = le64_to_cpu(src->max_val);
 623         dst->min_val    = le64_to_cpu(src->min_val);
 624         dst->samples    = le64_to_cpu(src->samples);
 625
 626         /*
 627          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 628          */
 629         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 630         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 631 }
 632
 633 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 634 {
 635         int i, j;
 636
 637         dst->error              = le32_to_cpu(src->error);
 638         dst->thread_number      = le32_to_cpu(src->thread_number);
 639         dst->groupid            = le32_to_cpu(src->groupid);
 640         dst->pid                = le32_to_cpu(src->pid);
 641         dst->members            = le32_to_cpu(src->members);
 642
 643         for (i = 0; i < 2; i++) {
 644                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 645                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 646                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 647                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 648         }
 649
 650         dst->usr_time           = le64_to_cpu(src->usr_time);
 651         dst->sys_time           = le64_to_cpu(src->sys_time);
 652         dst->ctx                = le64_to_cpu(src->ctx);
 653         dst->minf               = le64_to_cpu(src->minf);
 654         dst->majf               = le64_to_cpu(src->majf);
 655         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 656
 657         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 658                 fio_fp64_t *fps = &src->percentile_list[i];
 659                 fio_fp64_t *fpd = &dst->percentile_list[i];
 660
 661                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 662         }
 663
 664         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 665                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 666                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 667                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 668         }
 669
 670         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 671                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 672                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 673         }
 674
 675         for (i = 0; i < 2; i++)
 676                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 677                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 678
 679         for (i = 0; i < 3; i++) {
 680                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 681                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 682         }
 683
 684         dst->total_submit       = le64_to_cpu(src->total_submit);
 685         dst->total_complete     = le64_to_cpu(src->total_complete);
 686
 687         for (i = 0; i < 2; i++) {
 688                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 689                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 690         }
 691
 692         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 693         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 694         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 695         dst->first_error        = le32_to_cpu(src->first_error);
 696         dst->kb_base            = le32_to_cpu(src->kb_base);
 697 }
 698
 699 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 700 {
 701         int i;
 702
 703         for (i = 0; i < 2; i++) {
 704                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 705                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 706                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 707                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 708                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 709                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 710         }
 711
 712         dst->kb_base    = le32_to_cpu(src->kb_base);
 713         dst->groupid    = le32_to_cpu(src->groupid);
 714 }
 715
 716 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
 717 {
 718         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 719
 720         show_thread_status(&p->ts, &p->rs);
 721
 722         if (sum_stat_clients == 1)
 723                 return;
 724
 725         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 726         sum_group_stats(&client_gs, &p->rs);
 727
 728         client_ts.members++;
 729         client_ts.thread_number = p->ts.thread_number;
 730         client_ts.groupid = p->ts.groupid;
 731
 732         if (++sum_stat_nr == sum_stat_clients) {
 733                 strcpy(client_ts.name, "All clients");
 734                 show_thread_status(&client_ts, &client_gs);
 735         }
 736 }
 737
 738 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
 739 {
 740         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 741
 742         show_group_stats(gs);
 743 }
 744
 745 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
 746 {
 747         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
 748         const char *buf = (const char *) pdu->buf;
 749         const char *name;
 750         int fio_unused ret;
 751
 752         name = client->name ? client->name : client->hostname;
 753
 754         if (!client->skip_newline)
 755                 fprintf(f_out, "<%s> ", name);
 756         ret = fwrite(buf, pdu->buf_len, 1, f_out);
 757         fflush(f_out);
 758         client->skip_newline = strchr(buf, '\n') == NULL;
 759 }
 760
 761 static void convert_agg(struct disk_util_agg *agg)
 762 {
 763         int i;
 764
 765         for (i = 0; i < 2; i++) {
 766                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 767                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 768                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 769                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 770         }
 771
 772         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 773         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 774         agg->slavecount         = le32_to_cpu(agg->slavecount);
 775         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 776 }
 777
 778 static void convert_dus(struct disk_util_stat *dus)
 779 {
 780         int i;
 781
 782         for (i = 0; i < 2; i++) {
 783                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
 784                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
 785                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
 786                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
 787         }
 788
 789         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
 790         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
 791         dus->msec               = le64_to_cpu(dus->msec);
 792 }
 793
 794 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 795 {
 796         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 797
 798         if (!client->disk_stats_shown) {
 799                 client->disk_stats_shown = 1;
 800                 log_info("\nDisk stats (read/write):\n");
 801         }
 802
 803         print_disk_util(&du->dus, &du->agg, terse_output);
 804 }
 805
 806 static void convert_jobs_eta(struct jobs_eta *je)
 807 {
 808         int i;
 809
 810         je->nr_running          = le32_to_cpu(je->nr_running);
 811         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 812         je->nr_pending          = le32_to_cpu(je->nr_pending);
 813         je->files_open          = le32_to_cpu(je->files_open);
 814
 815         for (i = 0; i < 2; i++) {
 816                 je->m_rate[i]           = le32_to_cpu(je->m_rate[i]);
 817                 je->t_rate[i]           = le32_to_cpu(je->t_rate[i]);
 818                 je->m_iops[i]           = le32_to_cpu(je->m_iops[i]);
 819                 je->t_iops[i]           = le32_to_cpu(je->t_iops[i]);
 820                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 821                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 822         }
 823
 824         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 825         je->eta_sec             = le64_to_cpu(je->eta_sec);
 826         je->nr_threads          = le32_to_cpu(je->nr_threads);
 827 }
 828
 829 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 830 {
 831         int i;
 832
 833         dst->nr_running         += je->nr_running;
 834         dst->nr_ramp            += je->nr_ramp;
 835         dst->nr_pending         += je->nr_pending;
 836         dst->files_open         += je->files_open;
 837
 838         for (i = 0; i < 2; i++) {
 839                 dst->m_rate[i]  += je->m_rate[i];
 840                 dst->t_rate[i]  += je->t_rate[i];
 841                 dst->m_iops[i]  += je->m_iops[i];
 842                 dst->t_iops[i]  += je->t_iops[i];
 843                 dst->rate[i]    += je->rate[i];
 844                 dst->iops[i]    += je->iops[i];
 845         }
 846
 847         dst->elapsed_sec        += je->elapsed_sec;
 848
 849         if (je->eta_sec > dst->eta_sec)
 850                 dst->eta_sec = je->eta_sec;
 851
 852         dst->nr_threads         += je->nr_threads;
 853         /* we need to handle je->run_str too ... */
 854 }
 855
 856 void fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
 857 {
 858         if (!--eta->pending) {
 859                 eta_fn(&eta->eta);
 860                 free(eta);
 861         }
 862 }
 863
 864 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 865 {
 866         struct fio_net_int_cmd *icmd = NULL;
 867         struct flist_head *entry;
 868
 869         flist_for_each(entry, &client->cmd_list) {
 870                 icmd = flist_entry(entry, struct fio_net_int_cmd, list);
 871
 872                 if (cmd->tag == (uintptr_t) icmd)
 873                         break;
 874
 875                 icmd = NULL;
 876         }
 877
 878         if (!icmd) {
 879                 log_err("fio: client: unable to find matching tag\n");
 880                 return;
 881         }
 882
 883         flist_del(&icmd->list);
 884         cmd->tag = icmd->saved_tag;
 885         free(icmd);
 886 }
 887
 888 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 889 {
 890         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 891         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 892
 893         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 894
 895         assert(client->eta_in_flight == eta);
 896
 897         client->eta_in_flight = NULL;
 898         flist_del_init(&client->eta_list);
 899
 900         if (client->ops->jobs_eta)
 901                 client->ops->jobs_eta(client, je);
 902
 903         fio_client_sum_jobs_eta(&eta->eta, je);
 904         fio_client_dec_jobs_eta(eta, client->ops->eta);
 905 }
 906
 907 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 908 {
 909         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 910         const char *os, *arch;
 911         char bit[16];
 912
 913         os = fio_get_os_string(probe->os);
 914         if (!os)
 915                 os = "unknown";
 916
 917         arch = fio_get_arch_string(probe->arch);
 918         if (!arch)
 919                 os = "unknown";
 920
 921         sprintf(bit, "%d-bit", probe->bpp * 8);
 922
 923         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
 924                 probe->hostname, probe->bigendian, bit, os, arch,
 925                 probe->fio_major, probe->fio_minor, probe->fio_patch);
 926
 927         if (!client->name)
 928                 client->name = strdup((char *) probe->hostname);
 929 }
 930
 931 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
 932 {
 933         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 934
 935         client->state = Client_started;
 936         client->jobs = pdu->jobs;
 937 }
 938
 939 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
 940 {
 941         if (client->error)
 942                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
 943 }
 944
 945 static void convert_stop(struct fio_net_cmd *cmd)
 946 {
 947         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
 948
 949         pdu->error = le32_to_cpu(pdu->error);
 950 }
 951
 952 static void convert_text(struct fio_net_cmd *cmd)
 953 {
 954         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
 955
 956         pdu->level      = le32_to_cpu(pdu->level);
 957         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
 958         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
 959         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
 960 }
 961
 962 /*
 963  * This has been compressed on the server side, since it can be big.
 964  * Uncompress here.
 965  */
 966 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd)
 967 {
 968         struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
 969         struct cmd_iolog_pdu *ret;
 970         uint32_t nr_samples;
 971         unsigned long total;
 972         z_stream stream;
 973         void *p;
 974         int i;
 975
 976         stream.zalloc = Z_NULL;
 977         stream.zfree = Z_NULL;
 978         stream.opaque = Z_NULL;
 979         stream.avail_in = 0;
 980         stream.next_in = Z_NULL;
 981
 982         if (inflateInit(&stream) != Z_OK)
 983                 return NULL;
 984
 985         /*
 986          * Get header first, it's not compressed
 987          */
 988         nr_samples = le32_to_cpu(pdu->nr_samples);
 989
 990         total = nr_samples * sizeof(struct io_sample);
 991         ret = malloc(total + sizeof(*pdu));
 992         ret->thread_number = le32_to_cpu(pdu->thread_number);
 993         ret->nr_samples = nr_samples;
 994         ret->log_type = le32_to_cpu(pdu->log_type);
 995         strcpy((char *) ret->name, (char *) pdu->name);
 996
 997         p = (void *) ret + sizeof(*pdu);
 998
 999         stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1000         stream.next_in = (void *) pdu + sizeof(*pdu);
1001         while (stream.avail_in) {
1002                 unsigned int this_chunk = 65536;
1003                 unsigned int this_len;
1004                 int err;
1005
1006                 if (this_chunk > total)
1007                         this_chunk = total;
1008
1009                 stream.avail_out = this_chunk;
1010                 stream.next_out = p;
1011                 err = inflate(&stream, Z_NO_FLUSH);
1012                 /* may be Z_OK, or Z_STREAM_END */
1013                 if (err < 0) {
1014                         log_err("fio: inflate error %d\n", err);
1015                         free(ret);
1016                         ret = NULL;
1017                         goto out;
1018                 }
1019
1020                 this_len = this_chunk - stream.avail_out;
1021                 p += this_len;
1022                 total -= this_len;
1023         }
1024
1025         for (i = 0; i < ret->nr_samples; i++) {
1026                 struct io_sample *s = &ret->samples[i];
1027
1028                 s->time = le64_to_cpu(s->time);
1029                 s->val  = le64_to_cpu(s->val);
1030                 s->ddir = le32_to_cpu(s->ddir);
1031                 s->bs   = le32_to_cpu(s->bs);
1032         }
1033
1034 out:
1035         inflateEnd(&stream);
1036         return ret;
1037 }
1038
1039 int fio_handle_client(struct fio_client *client)
1040 {
1041         struct client_ops *ops = client->ops;
1042         struct fio_net_cmd *cmd;
1043
1044         dprint(FD_NET, "client: handle %s\n", client->hostname);
1045
1046         cmd = fio_net_recv_cmd(client->fd);
1047         if (!cmd)
1048                 return 0;
1049
1050         dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1051                 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1052
1053         switch (cmd->opcode) {
1054         case FIO_NET_CMD_QUIT:
1055                 if (ops->quit)
1056                         ops->quit(client, cmd);
1057                 remove_client(client);
1058                 free(cmd);
1059                 break;
1060         case FIO_NET_CMD_TEXT:
1061                 convert_text(cmd);
1062                 ops->text(client, cmd);
1063                 free(cmd);
1064                 break;
1065         case FIO_NET_CMD_DU: {
1066                 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1067
1068                 convert_dus(&du->dus);
1069                 convert_agg(&du->agg);
1070
1071                 ops->disk_util(client, cmd);
1072                 free(cmd);
1073                 break;
1074                 }
1075         case FIO_NET_CMD_TS: {
1076                 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1077
1078                 convert_ts(&p->ts, &p->ts);
1079                 convert_gs(&p->rs, &p->rs);
1080
1081                 ops->thread_status(client, cmd);
1082                 free(cmd);
1083                 break;
1084                 }
1085         case FIO_NET_CMD_GS: {
1086                 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1087
1088                 convert_gs(gs, gs);
1089
1090                 ops->group_stats(client, cmd);
1091                 free(cmd);
1092                 break;
1093                 }
1094         case FIO_NET_CMD_ETA: {
1095                 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1096
1097                 remove_reply_cmd(client, cmd);
1098                 convert_jobs_eta(je);
1099                 handle_eta(client, cmd);
1100                 free(cmd);
1101                 break;
1102                 }
1103         case FIO_NET_CMD_PROBE:
1104                 remove_reply_cmd(client, cmd);
1105                 ops->probe(client, cmd);
1106                 free(cmd);
1107                 break;
1108         case FIO_NET_CMD_SERVER_START:
1109                 client->state = Client_running;
1110                 if (ops->job_start)
1111                         ops->job_start(client, cmd);
1112                 free(cmd);
1113                 break;
1114         case FIO_NET_CMD_START: {
1115                 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1116
1117                 pdu->jobs = le32_to_cpu(pdu->jobs);
1118                 ops->start(client, cmd);
1119                 free(cmd);
1120                 break;
1121                 }
1122         case FIO_NET_CMD_STOP: {
1123                 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1124
1125                 convert_stop(cmd);
1126                 client->state = Client_stopped;
1127                 client->error = le32_to_cpu(pdu->error);
1128                 client->signal = le32_to_cpu(pdu->signal);
1129                 ops->stop(client, cmd);
1130                 free(cmd);
1131                 break;
1132                 }
1133         case FIO_NET_CMD_ADD_JOB:
1134                 if (ops->add_job)
1135                         ops->add_job(client, cmd);
1136                 free(cmd);
1137                 break;
1138         case FIO_NET_CMD_IOLOG:
1139                 if (ops->iolog) {
1140                         struct cmd_iolog_pdu *pdu;
1141
1142                         pdu = convert_iolog(cmd);
1143                         ops->iolog(client, pdu);
1144                 }
1145                 free(cmd);
1146                 break;
1147         default:
1148                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1149                 free(cmd);
1150                 break;
1151         }
1152
1153         return 1;
1154 }
1155
1156 static void request_client_etas(struct client_ops *ops)
1157 {
1158         struct fio_client *client;
1159         struct flist_head *entry;
1160         struct client_eta *eta;
1161         int skipped = 0;
1162
1163         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1164
1165         eta = malloc(sizeof(*eta));
1166         memset(&eta->eta, 0, sizeof(eta->eta));
1167         eta->pending = nr_clients;
1168
1169         flist_for_each(entry, &client_list) {
1170                 client = flist_entry(entry, struct fio_client, list);
1171
1172                 if (!flist_empty(&client->eta_list)) {
1173                         skipped++;
1174                         continue;
1175                 }
1176                 if (client->state != Client_running)
1177                         continue;
1178
1179                 assert(!client->eta_in_flight);
1180                 flist_add_tail(&client->eta_list, &eta_list);
1181                 client->eta_in_flight = eta;
1182                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1183                                         (uintptr_t) eta, &client->cmd_list);
1184         }
1185
1186         while (skipped--)
1187                 fio_client_dec_jobs_eta(eta, ops->eta);
1188
1189         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1190 }
1191
1192 static int client_check_cmd_timeout(struct fio_client *client,
1193                                     struct timeval *now)
1194 {
1195         struct fio_net_int_cmd *cmd;
1196         struct flist_head *entry, *tmp;
1197         int ret = 0;
1198
1199         flist_for_each_safe(entry, tmp, &client->cmd_list) {
1200                 cmd = flist_entry(entry, struct fio_net_int_cmd, list);
1201
1202                 if (mtime_since(&cmd->tv, now) < FIO_NET_CLIENT_TIMEOUT)
1203                         continue;
1204
1205                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1206                                                 fio_server_op(cmd->cmd.opcode));
1207                 flist_del(&cmd->list);
1208                 free(cmd);
1209                 ret = 1;
1210         }
1211
1212         return flist_empty(&client->cmd_list) && ret;
1213 }
1214
1215 static int fio_check_clients_timed_out(void)
1216 {
1217         struct fio_client *client;
1218         struct flist_head *entry, *tmp;
1219         struct timeval tv;
1220         int ret = 0;
1221
1222         gettimeofday(&tv, NULL);
1223
1224         flist_for_each_safe(entry, tmp, &client_list) {
1225                 client = flist_entry(entry, struct fio_client, list);
1226
1227                 if (flist_empty(&client->cmd_list))
1228                         continue;
1229
1230                 if (!client_check_cmd_timeout(client, &tv))
1231                         continue;
1232
1233                 if (client->ops->timed_out)
1234                         client->ops->timed_out(client);
1235                 else
1236                         log_err("fio: client %s timed out\n", client->hostname);
1237
1238                 remove_client(client);
1239                 ret = 1;
1240         }
1241
1242         return ret;
1243 }
1244
1245 int fio_handle_clients(struct client_ops *ops)
1246 {
1247         struct pollfd *pfds;
1248         int i, ret = 0, retval = 0;
1249
1250         gettimeofday(&eta_tv, NULL);
1251
1252         pfds = malloc(nr_clients * sizeof(struct pollfd));
1253
1254         sum_stat_clients = nr_clients;
1255         init_thread_stat(&client_ts);
1256         init_group_run_stat(&client_gs);
1257
1258         while (!exit_backend && nr_clients) {
1259                 struct flist_head *entry, *tmp;
1260                 struct fio_client *client;
1261
1262                 i = 0;
1263                 flist_for_each_safe(entry, tmp, &client_list) {
1264                         client = flist_entry(entry, struct fio_client, list);
1265
1266                         if (!client->sent_job && !client->ops->stay_connected &&
1267                             flist_empty(&client->cmd_list)) {
1268                                 remove_client(client);
1269                                 continue;
1270                         }
1271
1272                         pfds[i].fd = client->fd;
1273                         pfds[i].events = POLLIN;
1274                         i++;
1275                 }
1276
1277                 if (!nr_clients)
1278                         break;
1279
1280                 assert(i == nr_clients);
1281
1282                 do {
1283                         struct timeval tv;
1284
1285                         gettimeofday(&tv, NULL);
1286                         if (mtime_since(&eta_tv, &tv) >= ops->eta_msec) {
1287                                 request_client_etas(ops);
1288                                 memcpy(&eta_tv, &tv, sizeof(tv));
1289
1290                                 if (fio_check_clients_timed_out())
1291                                         break;
1292                         }
1293
1294                         ret = poll(pfds, nr_clients, 100);
1295                         if (ret < 0) {
1296                                 if (errno == EINTR)
1297                                         continue;
1298                                 log_err("fio: poll clients: %s\n", strerror(errno));
1299                                 break;
1300                         } else if (!ret)
1301                                 continue;
1302                 } while (ret <= 0);
1303
1304                 for (i = 0; i < nr_clients; i++) {
1305                         if (!(pfds[i].revents & POLLIN))
1306                                 continue;
1307
1308                         client = find_client_by_fd(pfds[i].fd);
1309                         if (!client) {
1310                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1311                                 continue;
1312                         }
1313                         if (!fio_handle_client(client)) {
1314                                 log_info("client: host=%s disconnected\n",
1315                                                 client->hostname);
1316                                 remove_client(client);
1317                                 retval = 1;
1318                         } else if (client->error)
1319                                 retval = 1;
1320                         fio_put_client(client);
1321                 }
1322         }
1323
1324         free(pfds);
1325         return retval;
1326 }