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, NULL, NULL);
 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, NULL, NULL);
 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 int fio_client_update_options(struct fio_client *client,
 621                               struct thread_options *o, uint64_t *tag)
 622 {
 623         struct cmd_add_job_pdu pdu;
 624
 625         pdu.thread_number = cpu_to_le32(client->thread_number);
 626         pdu.groupid = cpu_to_le32(client->groupid);
 627         convert_thread_options_to_net(&pdu.top, o);
 628
 629         return fio_net_send_cmd(client->fd, FIO_NET_CMD_UPDATE_JOB, &pdu, sizeof(pdu), tag, &client->cmd_list);
 630 }
 631
 632 static void convert_io_stat(struct io_stat *dst, struct io_stat *src)
 633 {
 634         dst->max_val    = le64_to_cpu(src->max_val);
 635         dst->min_val    = le64_to_cpu(src->min_val);
 636         dst->samples    = le64_to_cpu(src->samples);
 637
 638         /*
 639          * Floats arrive as IEEE 754 encoded uint64_t, convert back to double
 640          */
 641         dst->mean.u.f   = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i));
 642         dst->S.u.f      = fio_uint64_to_double(le64_to_cpu(dst->S.u.i));
 643 }
 644
 645 static void convert_ts(struct thread_stat *dst, struct thread_stat *src)
 646 {
 647         int i, j;
 648
 649         dst->error              = le32_to_cpu(src->error);
 650         dst->thread_number      = le32_to_cpu(src->thread_number);
 651         dst->groupid            = le32_to_cpu(src->groupid);
 652         dst->pid                = le32_to_cpu(src->pid);
 653         dst->members            = le32_to_cpu(src->members);
 654
 655         for (i = 0; i < 2; i++) {
 656                 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]);
 657                 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]);
 658                 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]);
 659                 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]);
 660         }
 661
 662         dst->usr_time           = le64_to_cpu(src->usr_time);
 663         dst->sys_time           = le64_to_cpu(src->sys_time);
 664         dst->ctx                = le64_to_cpu(src->ctx);
 665         dst->minf               = le64_to_cpu(src->minf);
 666         dst->majf               = le64_to_cpu(src->majf);
 667         dst->clat_percentiles   = le64_to_cpu(src->clat_percentiles);
 668
 669         for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
 670                 fio_fp64_t *fps = &src->percentile_list[i];
 671                 fio_fp64_t *fpd = &dst->percentile_list[i];
 672
 673                 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i));
 674         }
 675
 676         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
 677                 dst->io_u_map[i]        = le32_to_cpu(src->io_u_map[i]);
 678                 dst->io_u_submit[i]     = le32_to_cpu(src->io_u_submit[i]);
 679                 dst->io_u_complete[i]   = le32_to_cpu(src->io_u_complete[i]);
 680         }
 681
 682         for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
 683                 dst->io_u_lat_u[i]      = le32_to_cpu(src->io_u_lat_u[i]);
 684                 dst->io_u_lat_m[i]      = le32_to_cpu(src->io_u_lat_m[i]);
 685         }
 686
 687         for (i = 0; i < 2; i++)
 688                 for (j = 0; j < FIO_IO_U_PLAT_NR; j++)
 689                         dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]);
 690
 691         for (i = 0; i < 3; i++) {
 692                 dst->total_io_u[i]      = le64_to_cpu(src->total_io_u[i]);
 693                 dst->short_io_u[i]      = le64_to_cpu(src->short_io_u[i]);
 694         }
 695
 696         dst->total_submit       = le64_to_cpu(src->total_submit);
 697         dst->total_complete     = le64_to_cpu(src->total_complete);
 698
 699         for (i = 0; i < 2; i++) {
 700                 dst->io_bytes[i]        = le64_to_cpu(src->io_bytes[i]);
 701                 dst->runtime[i]         = le64_to_cpu(src->runtime[i]);
 702         }
 703
 704         dst->total_run_time     = le64_to_cpu(src->total_run_time);
 705         dst->continue_on_error  = le16_to_cpu(src->continue_on_error);
 706         dst->total_err_count    = le64_to_cpu(src->total_err_count);
 707         dst->first_error        = le32_to_cpu(src->first_error);
 708         dst->kb_base            = le32_to_cpu(src->kb_base);
 709 }
 710
 711 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src)
 712 {
 713         int i;
 714
 715         for (i = 0; i < 2; i++) {
 716                 dst->max_run[i]         = le64_to_cpu(src->max_run[i]);
 717                 dst->min_run[i]         = le64_to_cpu(src->min_run[i]);
 718                 dst->max_bw[i]          = le64_to_cpu(src->max_bw[i]);
 719                 dst->min_bw[i]          = le64_to_cpu(src->min_bw[i]);
 720                 dst->io_kb[i]           = le64_to_cpu(src->io_kb[i]);
 721                 dst->agg[i]             = le64_to_cpu(src->agg[i]);
 722         }
 723
 724         dst->kb_base    = le32_to_cpu(src->kb_base);
 725         dst->groupid    = le32_to_cpu(src->groupid);
 726 }
 727
 728 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd)
 729 {
 730         struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
 731
 732         show_thread_status(&p->ts, &p->rs);
 733
 734         if (sum_stat_clients == 1)
 735                 return;
 736
 737         sum_thread_stats(&client_ts, &p->ts, sum_stat_nr);
 738         sum_group_stats(&client_gs, &p->rs);
 739
 740         client_ts.members++;
 741         client_ts.thread_number = p->ts.thread_number;
 742         client_ts.groupid = p->ts.groupid;
 743
 744         if (++sum_stat_nr == sum_stat_clients) {
 745                 strcpy(client_ts.name, "All clients");
 746                 show_thread_status(&client_ts, &client_gs);
 747         }
 748 }
 749
 750 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd)
 751 {
 752         struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
 753
 754         show_group_stats(gs);
 755 }
 756
 757 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd)
 758 {
 759         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
 760         const char *buf = (const char *) pdu->buf;
 761         const char *name;
 762         int fio_unused ret;
 763
 764         name = client->name ? client->name : client->hostname;
 765
 766         if (!client->skip_newline)
 767                 fprintf(f_out, "<%s> ", name);
 768         ret = fwrite(buf, pdu->buf_len, 1, f_out);
 769         fflush(f_out);
 770         client->skip_newline = strchr(buf, '\n') == NULL;
 771 }
 772
 773 static void convert_agg(struct disk_util_agg *agg)
 774 {
 775         int i;
 776
 777         for (i = 0; i < 2; i++) {
 778                 agg->ios[i]     = le32_to_cpu(agg->ios[i]);
 779                 agg->merges[i]  = le32_to_cpu(agg->merges[i]);
 780                 agg->sectors[i] = le64_to_cpu(agg->sectors[i]);
 781                 agg->ticks[i]   = le32_to_cpu(agg->ticks[i]);
 782         }
 783
 784         agg->io_ticks           = le32_to_cpu(agg->io_ticks);
 785         agg->time_in_queue      = le32_to_cpu(agg->time_in_queue);
 786         agg->slavecount         = le32_to_cpu(agg->slavecount);
 787         agg->max_util.u.f       = fio_uint64_to_double(__le64_to_cpu(agg->max_util.u.i));
 788 }
 789
 790 static void convert_dus(struct disk_util_stat *dus)
 791 {
 792         int i;
 793
 794         for (i = 0; i < 2; i++) {
 795                 dus->ios[i]     = le32_to_cpu(dus->ios[i]);
 796                 dus->merges[i]  = le32_to_cpu(dus->merges[i]);
 797                 dus->sectors[i] = le64_to_cpu(dus->sectors[i]);
 798                 dus->ticks[i]   = le32_to_cpu(dus->ticks[i]);
 799         }
 800
 801         dus->io_ticks           = le32_to_cpu(dus->io_ticks);
 802         dus->time_in_queue      = le32_to_cpu(dus->time_in_queue);
 803         dus->msec               = le64_to_cpu(dus->msec);
 804 }
 805
 806 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd)
 807 {
 808         struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
 809
 810         if (!client->disk_stats_shown) {
 811                 client->disk_stats_shown = 1;
 812                 log_info("\nDisk stats (read/write):\n");
 813         }
 814
 815         print_disk_util(&du->dus, &du->agg, terse_output);
 816 }
 817
 818 static void convert_jobs_eta(struct jobs_eta *je)
 819 {
 820         int i;
 821
 822         je->nr_running          = le32_to_cpu(je->nr_running);
 823         je->nr_ramp             = le32_to_cpu(je->nr_ramp);
 824         je->nr_pending          = le32_to_cpu(je->nr_pending);
 825         je->files_open          = le32_to_cpu(je->files_open);
 826
 827         for (i = 0; i < 2; i++) {
 828                 je->m_rate[i]           = le32_to_cpu(je->m_rate[i]);
 829                 je->t_rate[i]           = le32_to_cpu(je->t_rate[i]);
 830                 je->m_iops[i]           = le32_to_cpu(je->m_iops[i]);
 831                 je->t_iops[i]           = le32_to_cpu(je->t_iops[i]);
 832                 je->rate[i]     = le32_to_cpu(je->rate[i]);
 833                 je->iops[i]     = le32_to_cpu(je->iops[i]);
 834         }
 835
 836         je->elapsed_sec         = le64_to_cpu(je->elapsed_sec);
 837         je->eta_sec             = le64_to_cpu(je->eta_sec);
 838         je->nr_threads          = le32_to_cpu(je->nr_threads);
 839 }
 840
 841 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je)
 842 {
 843         int i;
 844
 845         dst->nr_running         += je->nr_running;
 846         dst->nr_ramp            += je->nr_ramp;
 847         dst->nr_pending         += je->nr_pending;
 848         dst->files_open         += je->files_open;
 849
 850         for (i = 0; i < 2; i++) {
 851                 dst->m_rate[i]  += je->m_rate[i];
 852                 dst->t_rate[i]  += je->t_rate[i];
 853                 dst->m_iops[i]  += je->m_iops[i];
 854                 dst->t_iops[i]  += je->t_iops[i];
 855                 dst->rate[i]    += je->rate[i];
 856                 dst->iops[i]    += je->iops[i];
 857         }
 858
 859         dst->elapsed_sec        += je->elapsed_sec;
 860
 861         if (je->eta_sec > dst->eta_sec)
 862                 dst->eta_sec = je->eta_sec;
 863
 864         dst->nr_threads         += je->nr_threads;
 865         /* we need to handle je->run_str too ... */
 866 }
 867
 868 void fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn)
 869 {
 870         if (!--eta->pending) {
 871                 eta_fn(&eta->eta);
 872                 free(eta);
 873         }
 874 }
 875
 876 static void remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd)
 877 {
 878         struct fio_net_cmd_reply *reply = NULL;
 879         struct flist_head *entry;
 880
 881         flist_for_each(entry, &client->cmd_list) {
 882                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
 883
 884                 if (cmd->tag == (uintptr_t) reply)
 885                         break;
 886
 887                 reply = NULL;
 888         }
 889
 890         if (!reply) {
 891                 log_err("fio: client: unable to find matching tag (%lx)\n", cmd->tag);
 892                 return;
 893         }
 894
 895         flist_del(&reply->list);
 896         cmd->tag = reply->saved_tag;
 897         free(reply);
 898 }
 899
 900 int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag)
 901 {
 902         do {
 903                 struct fio_net_cmd_reply *reply = NULL;
 904                 struct flist_head *entry;
 905
 906                 flist_for_each(entry, &client->cmd_list) {
 907                         reply = flist_entry(entry, struct fio_net_cmd_reply, list);
 908
 909                         if (tag == (uintptr_t) reply)
 910                                 break;
 911
 912                         reply = NULL;
 913                 }
 914
 915                 if (!reply)
 916                         break;
 917
 918                 usleep(1000);
 919         } while (1);
 920
 921         return 0;
 922 }
 923
 924 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd)
 925 {
 926         struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
 927         struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag;
 928
 929         dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending);
 930
 931         assert(client->eta_in_flight == eta);
 932
 933         client->eta_in_flight = NULL;
 934         flist_del_init(&client->eta_list);
 935
 936         if (client->ops->jobs_eta)
 937                 client->ops->jobs_eta(client, je);
 938
 939         fio_client_sum_jobs_eta(&eta->eta, je);
 940         fio_client_dec_jobs_eta(eta, client->ops->eta);
 941 }
 942
 943 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd)
 944 {
 945         struct cmd_probe_pdu *probe = (struct cmd_probe_pdu *) cmd->payload;
 946         const char *os, *arch;
 947         char bit[16];
 948
 949         os = fio_get_os_string(probe->os);
 950         if (!os)
 951                 os = "unknown";
 952
 953         arch = fio_get_arch_string(probe->arch);
 954         if (!arch)
 955                 os = "unknown";
 956
 957         sprintf(bit, "%d-bit", probe->bpp * 8);
 958
 959         log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%u.%u.%u\n",
 960                 probe->hostname, probe->bigendian, bit, os, arch,
 961                 probe->fio_major, probe->fio_minor, probe->fio_patch);
 962
 963         if (!client->name)
 964                 client->name = strdup((char *) probe->hostname);
 965 }
 966
 967 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd)
 968 {
 969         struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
 970
 971         client->state = Client_started;
 972         client->jobs = pdu->jobs;
 973 }
 974
 975 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd)
 976 {
 977         if (client->error)
 978                 log_info("client <%s>: exited with error %d\n", client->hostname, client->error);
 979 }
 980
 981 static void convert_stop(struct fio_net_cmd *cmd)
 982 {
 983         struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
 984
 985         pdu->error = le32_to_cpu(pdu->error);
 986 }
 987
 988 static void convert_text(struct fio_net_cmd *cmd)
 989 {
 990         struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload;
 991
 992         pdu->level      = le32_to_cpu(pdu->level);
 993         pdu->buf_len    = le32_to_cpu(pdu->buf_len);
 994         pdu->log_sec    = le64_to_cpu(pdu->log_sec);
 995         pdu->log_usec   = le64_to_cpu(pdu->log_usec);
 996 }
 997
 998 /*
 999  * This has been compressed on the server side, since it can be big.
1000  * Uncompress here.
1001  */
1002 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd)
1003 {
1004         struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload;
1005         struct cmd_iolog_pdu *ret;
1006         uint32_t nr_samples;
1007         unsigned long total;
1008         z_stream stream;
1009         void *p;
1010         int i;
1011
1012         stream.zalloc = Z_NULL;
1013         stream.zfree = Z_NULL;
1014         stream.opaque = Z_NULL;
1015         stream.avail_in = 0;
1016         stream.next_in = Z_NULL;
1017
1018         if (inflateInit(&stream) != Z_OK)
1019                 return NULL;
1020
1021         /*
1022          * Get header first, it's not compressed
1023          */
1024         nr_samples = le32_to_cpu(pdu->nr_samples);
1025
1026         total = nr_samples * sizeof(struct io_sample);
1027         ret = malloc(total + sizeof(*pdu));
1028         ret->thread_number = le32_to_cpu(pdu->thread_number);
1029         ret->nr_samples = nr_samples;
1030         ret->log_type = le32_to_cpu(pdu->log_type);
1031         strcpy((char *) ret->name, (char *) pdu->name);
1032
1033         p = (void *) ret + sizeof(*pdu);
1034
1035         stream.avail_in = cmd->pdu_len - sizeof(*pdu);
1036         stream.next_in = (void *) pdu + sizeof(*pdu);
1037         while (stream.avail_in) {
1038                 unsigned int this_chunk = 65536;
1039                 unsigned int this_len;
1040                 int err;
1041
1042                 if (this_chunk > total)
1043                         this_chunk = total;
1044
1045                 stream.avail_out = this_chunk;
1046                 stream.next_out = p;
1047                 err = inflate(&stream, Z_NO_FLUSH);
1048                 /* may be Z_OK, or Z_STREAM_END */
1049                 if (err < 0) {
1050                         log_err("fio: inflate error %d\n", err);
1051                         free(ret);
1052                         ret = NULL;
1053                         goto out;
1054                 }
1055
1056                 this_len = this_chunk - stream.avail_out;
1057                 p += this_len;
1058                 total -= this_len;
1059         }
1060
1061         for (i = 0; i < ret->nr_samples; i++) {
1062                 struct io_sample *s = &ret->samples[i];
1063
1064                 s->time = le64_to_cpu(s->time);
1065                 s->val  = le64_to_cpu(s->val);
1066                 s->ddir = le32_to_cpu(s->ddir);
1067                 s->bs   = le32_to_cpu(s->bs);
1068         }
1069
1070 out:
1071         inflateEnd(&stream);
1072         return ret;
1073 }
1074
1075 int fio_handle_client(struct fio_client *client)
1076 {
1077         struct client_ops *ops = client->ops;
1078         struct fio_net_cmd *cmd;
1079
1080         dprint(FD_NET, "client: handle %s\n", client->hostname);
1081
1082         cmd = fio_net_recv_cmd(client->fd);
1083         if (!cmd)
1084                 return 0;
1085
1086         dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n",
1087                 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len);
1088
1089         switch (cmd->opcode) {
1090         case FIO_NET_CMD_QUIT:
1091                 if (ops->quit)
1092                         ops->quit(client, cmd);
1093                 remove_client(client);
1094                 free(cmd);
1095                 break;
1096         case FIO_NET_CMD_TEXT:
1097                 convert_text(cmd);
1098                 ops->text(client, cmd);
1099                 free(cmd);
1100                 break;
1101         case FIO_NET_CMD_DU: {
1102                 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload;
1103
1104                 convert_dus(&du->dus);
1105                 convert_agg(&du->agg);
1106
1107                 ops->disk_util(client, cmd);
1108                 free(cmd);
1109                 break;
1110                 }
1111         case FIO_NET_CMD_TS: {
1112                 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload;
1113
1114                 convert_ts(&p->ts, &p->ts);
1115                 convert_gs(&p->rs, &p->rs);
1116
1117                 ops->thread_status(client, cmd);
1118                 free(cmd);
1119                 break;
1120                 }
1121         case FIO_NET_CMD_GS: {
1122                 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload;
1123
1124                 convert_gs(gs, gs);
1125
1126                 ops->group_stats(client, cmd);
1127                 free(cmd);
1128                 break;
1129                 }
1130         case FIO_NET_CMD_ETA: {
1131                 struct jobs_eta *je = (struct jobs_eta *) cmd->payload;
1132
1133                 remove_reply_cmd(client, cmd);
1134                 convert_jobs_eta(je);
1135                 handle_eta(client, cmd);
1136                 free(cmd);
1137                 break;
1138                 }
1139         case FIO_NET_CMD_PROBE:
1140                 remove_reply_cmd(client, cmd);
1141                 ops->probe(client, cmd);
1142                 free(cmd);
1143                 break;
1144         case FIO_NET_CMD_SERVER_START:
1145                 client->state = Client_running;
1146                 if (ops->job_start)
1147                         ops->job_start(client, cmd);
1148                 free(cmd);
1149                 break;
1150         case FIO_NET_CMD_START: {
1151                 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload;
1152
1153                 pdu->jobs = le32_to_cpu(pdu->jobs);
1154                 ops->start(client, cmd);
1155                 free(cmd);
1156                 break;
1157                 }
1158         case FIO_NET_CMD_STOP: {
1159                 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload;
1160
1161                 convert_stop(cmd);
1162                 client->state = Client_stopped;
1163                 client->error = le32_to_cpu(pdu->error);
1164                 client->signal = le32_to_cpu(pdu->signal);
1165                 ops->stop(client, cmd);
1166                 free(cmd);
1167                 break;
1168                 }
1169         case FIO_NET_CMD_ADD_JOB: {
1170                 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload;
1171
1172                 client->thread_number = le32_to_cpu(pdu->thread_number);
1173                 client->groupid = le32_to_cpu(pdu->groupid);
1174
1175                 if (ops->add_job)
1176                         ops->add_job(client, cmd);
1177                 free(cmd);
1178                 break;
1179                 }
1180         case FIO_NET_CMD_IOLOG:
1181                 if (ops->iolog) {
1182                         struct cmd_iolog_pdu *pdu;
1183
1184                         pdu = convert_iolog(cmd);
1185                         ops->iolog(client, pdu);
1186                 }
1187                 free(cmd);
1188                 break;
1189         case FIO_NET_CMD_UPDATE_JOB:
1190                 remove_reply_cmd(client, cmd);
1191                 ops->update_job(client, cmd);
1192                 free(cmd);
1193                 break;
1194         default:
1195                 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode));
1196                 free(cmd);
1197                 break;
1198         }
1199
1200         return 1;
1201 }
1202
1203 static void request_client_etas(struct client_ops *ops)
1204 {
1205         struct fio_client *client;
1206         struct flist_head *entry;
1207         struct client_eta *eta;
1208         int skipped = 0;
1209
1210         dprint(FD_NET, "client: request eta (%d)\n", nr_clients);
1211
1212         eta = malloc(sizeof(*eta));
1213         memset(&eta->eta, 0, sizeof(eta->eta));
1214         eta->pending = nr_clients;
1215
1216         flist_for_each(entry, &client_list) {
1217                 client = flist_entry(entry, struct fio_client, list);
1218
1219                 if (!flist_empty(&client->eta_list)) {
1220                         skipped++;
1221                         continue;
1222                 }
1223                 if (client->state != Client_running)
1224                         continue;
1225
1226                 assert(!client->eta_in_flight);
1227                 flist_add_tail(&client->eta_list, &eta_list);
1228                 client->eta_in_flight = eta;
1229                 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA,
1230                                         (uintptr_t) eta, &client->cmd_list);
1231         }
1232
1233         while (skipped--)
1234                 fio_client_dec_jobs_eta(eta, ops->eta);
1235
1236         dprint(FD_NET, "client: requested eta tag %p\n", eta);
1237 }
1238
1239 static int client_check_cmd_timeout(struct fio_client *client,
1240                                     struct timeval *now)
1241 {
1242         struct fio_net_cmd_reply *reply;
1243         struct flist_head *entry, *tmp;
1244         int ret = 0;
1245
1246         flist_for_each_safe(entry, tmp, &client->cmd_list) {
1247                 reply = flist_entry(entry, struct fio_net_cmd_reply, list);
1248
1249                 if (mtime_since(&reply->tv, now) < FIO_NET_CLIENT_TIMEOUT)
1250                         continue;
1251
1252                 log_err("fio: client %s, timeout on cmd %s\n", client->hostname,
1253                                                 fio_server_op(reply->opcode));
1254                 flist_del(&reply->list);
1255                 free(reply);
1256                 ret = 1;
1257         }
1258
1259         return flist_empty(&client->cmd_list) && ret;
1260 }
1261
1262 static int fio_check_clients_timed_out(void)
1263 {
1264         struct fio_client *client;
1265         struct flist_head *entry, *tmp;
1266         struct timeval tv;
1267         int ret = 0;
1268
1269         gettimeofday(&tv, NULL);
1270
1271         flist_for_each_safe(entry, tmp, &client_list) {
1272                 client = flist_entry(entry, struct fio_client, list);
1273
1274                 if (flist_empty(&client->cmd_list))
1275                         continue;
1276
1277                 if (!client_check_cmd_timeout(client, &tv))
1278                         continue;
1279
1280                 if (client->ops->timed_out)
1281                         client->ops->timed_out(client);
1282                 else
1283                         log_err("fio: client %s timed out\n", client->hostname);
1284
1285                 remove_client(client);
1286                 ret = 1;
1287         }
1288
1289         return ret;
1290 }
1291
1292 int fio_handle_clients(struct client_ops *ops)
1293 {
1294         struct pollfd *pfds;
1295         int i, ret = 0, retval = 0;
1296
1297         gettimeofday(&eta_tv, NULL);
1298
1299         pfds = malloc(nr_clients * sizeof(struct pollfd));
1300
1301         sum_stat_clients = nr_clients;
1302         init_thread_stat(&client_ts);
1303         init_group_run_stat(&client_gs);
1304
1305         while (!exit_backend && nr_clients) {
1306                 struct flist_head *entry, *tmp;
1307                 struct fio_client *client;
1308
1309                 i = 0;
1310                 flist_for_each_safe(entry, tmp, &client_list) {
1311                         client = flist_entry(entry, struct fio_client, list);
1312
1313                         if (!client->sent_job && !client->ops->stay_connected &&
1314                             flist_empty(&client->cmd_list)) {
1315                                 remove_client(client);
1316                                 continue;
1317                         }
1318
1319                         pfds[i].fd = client->fd;
1320                         pfds[i].events = POLLIN;
1321                         i++;
1322                 }
1323
1324                 if (!nr_clients)
1325                         break;
1326
1327                 assert(i == nr_clients);
1328
1329                 do {
1330                         struct timeval tv;
1331
1332                         gettimeofday(&tv, NULL);
1333                         if (mtime_since(&eta_tv, &tv) >= ops->eta_msec) {
1334                                 request_client_etas(ops);
1335                                 memcpy(&eta_tv, &tv, sizeof(tv));
1336
1337                                 if (fio_check_clients_timed_out())
1338                                         break;
1339                         }
1340
1341                         ret = poll(pfds, nr_clients, ops->eta_msec);
1342                         if (ret < 0) {
1343                                 if (errno == EINTR)
1344                                         continue;
1345                                 log_err("fio: poll clients: %s\n", strerror(errno));
1346                                 break;
1347                         } else if (!ret)
1348                                 continue;
1349                 } while (ret <= 0);
1350
1351                 for (i = 0; i < nr_clients; i++) {
1352                         if (!(pfds[i].revents & POLLIN))
1353                                 continue;
1354
1355                         client = find_client_by_fd(pfds[i].fd);
1356                         if (!client) {
1357                                 log_err("fio: unknown client fd %d\n", pfds[i].fd);
1358                                 continue;
1359                         }
1360                         if (!fio_handle_client(client)) {
1361                                 log_info("client: host=%s disconnected\n",
1362                                                 client->hostname);
1363                                 remove_client(client);
1364                                 retval = 1;
1365                         } else if (client->error)
1366                                 retval = 1;
1367                         fio_put_client(client);
1368                 }
1369         }
1370
1371         free(pfds);
1372         return retval;
1373 }