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 #ifdef CONFIG_ZLIB 18 #include <zlib.h> 19 #endif 20 21 #include "fio.h" 22 #include "client.h" 23 #include "server.h" 24 #include "flist.h" 25 #include "hash.h" 26 #include "verify.h" 27 28 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd); 29 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd); 30 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd); 31 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd); 32 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd); 33 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd); 34 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd); 35 36 static void convert_text(struct fio_net_cmd *cmd); 37 38 struct client_ops fio_client_ops = { 39 .text = handle_text, 40 .disk_util = handle_du, 41 .thread_status = handle_ts, 42 .group_stats = handle_gs, 43 .stop = handle_stop, 44 .start = handle_start, 45 .eta = display_thread_status, 46 .probe = handle_probe, 47 .eta_msec = FIO_CLIENT_DEF_ETA_MSEC, 48 .client_type = FIO_CLIENT_TYPE_CLI, 49 }; 50 51 static struct timeval eta_tv; 52 53 static FLIST_HEAD(client_list); 54 static FLIST_HEAD(eta_list); 55 56 static FLIST_HEAD(arg_list); 57 58 struct thread_stat client_ts; 59 struct group_run_stats client_gs; 60 int sum_stat_clients; 61 62 static int sum_stat_nr; 63 static struct json_object *root = NULL; 64 static struct json_object *job_opt_object = NULL; 65 static struct json_array *clients_array = NULL; 66 static struct json_array *du_array = NULL; 67 68 static int error_clients; 69 70 #define FIO_CLIENT_HASH_BITS 7 71 #define FIO_CLIENT_HASH_SZ (1 << FIO_CLIENT_HASH_BITS) 72 #define FIO_CLIENT_HASH_MASK (FIO_CLIENT_HASH_SZ - 1) 73 static struct flist_head client_hash[FIO_CLIENT_HASH_SZ]; 74 75 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *, bool *); 76 77 static void fio_client_add_hash(struct fio_client *client) 78 { 79 int bucket = hash_long(client->fd, FIO_CLIENT_HASH_BITS); 80 81 bucket &= FIO_CLIENT_HASH_MASK; 82 flist_add(&client->hash_list, &client_hash[bucket]); 83 } 84 85 static void fio_client_remove_hash(struct fio_client *client) 86 { 87 if (!flist_empty(&client->hash_list)) 88 flist_del_init(&client->hash_list); 89 } 90 91 static void fio_init fio_client_hash_init(void) 92 { 93 int i; 94 95 for (i = 0; i < FIO_CLIENT_HASH_SZ; i++) 96 INIT_FLIST_HEAD(&client_hash[i]); 97 } 98 99 static int read_data(int fd, void *data, size_t size) 100 { 101 ssize_t ret; 102 103 while (size) { 104 ret = read(fd, data, size); 105 if (ret < 0) { 106 if (errno == EAGAIN || errno == EINTR) 107 continue; 108 break; 109 } else if (!ret) 110 break; 111 else { 112 data += ret; 113 size -= ret; 114 } 115 } 116 117 if (size) 118 return EAGAIN; 119 120 return 0; 121 } 122 123 static void fio_client_json_init(void) 124 { 125 char time_buf[32]; 126 time_t time_p; 127 128 if (!(output_format & FIO_OUTPUT_JSON)) 129 return; 130 131 time(&time_p); 132 os_ctime_r((const time_t *) &time_p, time_buf, sizeof(time_buf)); 133 time_buf[strlen(time_buf) - 1] = '\0'; 134 135 root = json_create_object(); 136 json_object_add_value_string(root, "fio version", fio_version_string); 137 json_object_add_value_int(root, "timestamp", time_p); 138 json_object_add_value_string(root, "time", time_buf); 139 140 job_opt_object = json_create_object(); 141 json_object_add_value_object(root, "global options", job_opt_object); 142 clients_array = json_create_array(); 143 json_object_add_value_array(root, "client_stats", clients_array); 144 du_array = json_create_array(); 145 json_object_add_value_array(root, "disk_util", du_array); 146 } 147 148 static void fio_client_json_fini(void) 149 { 150 if (!(output_format & FIO_OUTPUT_JSON)) 151 return; 152 153 log_info("\n"); 154 json_print_object(root, NULL); 155 log_info("\n"); 156 json_free_object(root); 157 root = NULL; 158 clients_array = NULL; 159 du_array = NULL; 160 } 161 162 static struct fio_client *find_client_by_fd(int fd) 163 { 164 int bucket = hash_long(fd, FIO_CLIENT_HASH_BITS) & FIO_CLIENT_HASH_MASK; 165 struct fio_client *client; 166 struct flist_head *entry; 167 168 flist_for_each(entry, &client_hash[bucket]) { 169 client = flist_entry(entry, struct fio_client, hash_list); 170 171 if (client->fd == fd) { 172 client->refs++; 173 return client; 174 } 175 } 176 177 return NULL; 178 } 179 180 void fio_put_client(struct fio_client *client) 181 { 182 if (--client->refs) 183 return; 184 185 free(client->hostname); 186 if (client->argv) 187 free(client->argv); 188 if (client->name) 189 free(client->name); 190 while (client->nr_files) { 191 struct client_file *cf = &client->files[--client->nr_files]; 192 193 free(cf->file); 194 } 195 if (client->files) 196 free(client->files); 197 if (client->opt_lists) 198 free(client->opt_lists); 199 200 if (!client->did_stat) 201 sum_stat_clients--; 202 203 if (client->error) 204 error_clients++; 205 206 free(client); 207 } 208 209 static int fio_client_dec_jobs_eta(struct client_eta *eta, client_eta_op eta_fn) 210 { 211 if (!--eta->pending) { 212 eta_fn(&eta->eta); 213 free(eta); 214 return 0; 215 } 216 217 return 1; 218 } 219 220 static void fio_drain_client_text(struct fio_client *client) 221 { 222 do { 223 struct fio_net_cmd *cmd; 224 225 cmd = fio_net_recv_cmd(client->fd, false); 226 if (!cmd) 227 break; 228 229 if (cmd->opcode == FIO_NET_CMD_TEXT) { 230 convert_text(cmd); 231 client->ops->text(client, cmd); 232 } 233 234 free(cmd); 235 } while (1); 236 } 237 238 static void remove_client(struct fio_client *client) 239 { 240 assert(client->refs); 241 242 dprint(FD_NET, "client: removed <%s>\n", client->hostname); 243 244 fio_drain_client_text(client); 245 246 if (!flist_empty(&client->list)) 247 flist_del_init(&client->list); 248 249 fio_client_remove_hash(client); 250 251 if (!flist_empty(&client->eta_list)) { 252 flist_del_init(&client->eta_list); 253 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta); 254 } 255 256 close(client->fd); 257 client->fd = -1; 258 259 if (client->ops->removed) 260 client->ops->removed(client); 261 262 nr_clients--; 263 fio_put_client(client); 264 } 265 266 struct fio_client *fio_get_client(struct fio_client *client) 267 { 268 client->refs++; 269 return client; 270 } 271 272 static void __fio_client_add_cmd_option(struct fio_client *client, 273 const char *opt) 274 { 275 int index; 276 277 index = client->argc++; 278 client->argv = realloc(client->argv, sizeof(char *) * client->argc); 279 client->argv[index] = strdup(opt); 280 dprint(FD_NET, "client: add cmd %d: %s\n", index, opt); 281 } 282 283 void fio_client_add_cmd_option(void *cookie, const char *opt) 284 { 285 struct fio_client *client = cookie; 286 struct flist_head *entry; 287 288 if (!client || !opt) 289 return; 290 291 __fio_client_add_cmd_option(client, opt); 292 293 /* 294 * Duplicate arguments to shared client group 295 */ 296 flist_for_each(entry, &arg_list) { 297 client = flist_entry(entry, struct fio_client, arg_list); 298 299 __fio_client_add_cmd_option(client, opt); 300 } 301 } 302 303 struct fio_client *fio_client_add_explicit(struct client_ops *ops, 304 const char *hostname, int type, 305 int port) 306 { 307 struct fio_client *client; 308 309 client = malloc(sizeof(*client)); 310 memset(client, 0, sizeof(*client)); 311 312 INIT_FLIST_HEAD(&client->list); 313 INIT_FLIST_HEAD(&client->hash_list); 314 INIT_FLIST_HEAD(&client->arg_list); 315 INIT_FLIST_HEAD(&client->eta_list); 316 INIT_FLIST_HEAD(&client->cmd_list); 317 318 client->hostname = strdup(hostname); 319 320 if (type == Fio_client_socket) 321 client->is_sock = 1; 322 else { 323 int ipv6; 324 325 ipv6 = type == Fio_client_ipv6; 326 if (fio_server_parse_host(hostname, ipv6, 327 &client->addr.sin_addr, 328 &client->addr6.sin6_addr)) 329 goto err; 330 331 client->port = port; 332 } 333 334 client->fd = -1; 335 client->ops = ops; 336 client->refs = 1; 337 client->type = ops->client_type; 338 339 __fio_client_add_cmd_option(client, "fio"); 340 341 flist_add(&client->list, &client_list); 342 nr_clients++; 343 dprint(FD_NET, "client: added <%s>\n", client->hostname); 344 return client; 345 err: 346 free(client); 347 return NULL; 348 } 349 350 int fio_client_add_ini_file(void *cookie, const char *ini_file, bool remote) 351 { 352 struct fio_client *client = cookie; 353 struct client_file *cf; 354 size_t new_size; 355 void *new_files; 356 357 if (!client) 358 return 1; 359 360 dprint(FD_NET, "client <%s>: add ini %s\n", client->hostname, ini_file); 361 362 new_size = (client->nr_files + 1) * sizeof(struct client_file); 363 new_files = realloc(client->files, new_size); 364 if (!new_files) 365 return 1; 366 367 client->files = new_files; 368 cf = &client->files[client->nr_files]; 369 cf->file = strdup(ini_file); 370 cf->remote = remote; 371 client->nr_files++; 372 return 0; 373 } 374 375 int fio_client_add(struct client_ops *ops, const char *hostname, void **cookie) 376 { 377 struct fio_client *existing = *cookie; 378 struct fio_client *client; 379 380 if (existing) { 381 /* 382 * We always add our "exec" name as the option, hence 1 383 * means empty. 384 */ 385 if (existing->argc == 1) 386 flist_add_tail(&existing->arg_list, &arg_list); 387 else { 388 while (!flist_empty(&arg_list)) 389 flist_del_init(arg_list.next); 390 } 391 } 392 393 client = malloc(sizeof(*client)); 394 memset(client, 0, sizeof(*client)); 395 396 INIT_FLIST_HEAD(&client->list); 397 INIT_FLIST_HEAD(&client->hash_list); 398 INIT_FLIST_HEAD(&client->arg_list); 399 INIT_FLIST_HEAD(&client->eta_list); 400 INIT_FLIST_HEAD(&client->cmd_list); 401 402 if (fio_server_parse_string(hostname, &client->hostname, 403 &client->is_sock, &client->port, 404 &client->addr.sin_addr, 405 &client->addr6.sin6_addr, 406 &client->ipv6)) 407 return -1; 408 409 client->fd = -1; 410 client->ops = ops; 411 client->refs = 1; 412 client->type = ops->client_type; 413 414 __fio_client_add_cmd_option(client, "fio"); 415 416 flist_add(&client->list, &client_list); 417 nr_clients++; 418 dprint(FD_NET, "client: added <%s>\n", client->hostname); 419 *cookie = client; 420 return 0; 421 } 422 423 static const char *server_name(struct fio_client *client, char *buf, 424 size_t bufsize) 425 { 426 const char *from; 427 428 if (client->ipv6) 429 from = inet_ntop(AF_INET6, (struct sockaddr *) &client->addr6.sin6_addr, buf, bufsize); 430 else if (client->is_sock) 431 from = "sock"; 432 else 433 from = inet_ntop(AF_INET, (struct sockaddr *) &client->addr.sin_addr, buf, bufsize); 434 435 return from; 436 } 437 438 static void probe_client(struct fio_client *client) 439 { 440 struct cmd_client_probe_pdu pdu; 441 const char *sname; 442 uint64_t tag; 443 char buf[64]; 444 445 dprint(FD_NET, "client: send probe\n"); 446 447 #ifdef CONFIG_ZLIB 448 pdu.flags = __le64_to_cpu(FIO_PROBE_FLAG_ZLIB); 449 #else 450 pdu.flags = 0; 451 #endif 452 453 sname = server_name(client, buf, sizeof(buf)); 454 memset(pdu.server, 0, sizeof(pdu.server)); 455 strncpy((char *) pdu.server, sname, sizeof(pdu.server) - 1); 456 457 fio_net_send_cmd(client->fd, FIO_NET_CMD_PROBE, &pdu, sizeof(pdu), &tag, &client->cmd_list); 458 } 459 460 static int fio_client_connect_ip(struct fio_client *client) 461 { 462 struct sockaddr *addr; 463 socklen_t socklen; 464 int fd, domain; 465 466 if (client->ipv6) { 467 client->addr6.sin6_family = AF_INET6; 468 client->addr6.sin6_port = htons(client->port); 469 domain = AF_INET6; 470 addr = (struct sockaddr *) &client->addr6; 471 socklen = sizeof(client->addr6); 472 } else { 473 client->addr.sin_family = AF_INET; 474 client->addr.sin_port = htons(client->port); 475 domain = AF_INET; 476 addr = (struct sockaddr *) &client->addr; 477 socklen = sizeof(client->addr); 478 } 479 480 fd = socket(domain, SOCK_STREAM, 0); 481 if (fd < 0) { 482 int ret = -errno; 483 484 log_err("fio: socket: %s\n", strerror(errno)); 485 return ret; 486 } 487 488 if (connect(fd, addr, socklen) < 0) { 489 int ret = -errno; 490 491 log_err("fio: connect: %s\n", strerror(errno)); 492 log_err("fio: failed to connect to %s:%u\n", client->hostname, 493 client->port); 494 close(fd); 495 return ret; 496 } 497 498 return fd; 499 } 500 501 static int fio_client_connect_sock(struct fio_client *client) 502 { 503 struct sockaddr_un *addr = &client->addr_un; 504 socklen_t len; 505 int fd; 506 507 memset(addr, 0, sizeof(*addr)); 508 addr->sun_family = AF_UNIX; 509 strncpy(addr->sun_path, client->hostname, sizeof(addr->sun_path) - 1); 510 511 fd = socket(AF_UNIX, SOCK_STREAM, 0); 512 if (fd < 0) { 513 int ret = -errno; 514 515 log_err("fio: socket: %s\n", strerror(errno)); 516 return ret; 517 } 518 519 len = sizeof(addr->sun_family) + strlen(addr->sun_path) + 1; 520 if (connect(fd, (struct sockaddr *) addr, len) < 0) { 521 int ret = -errno; 522 523 log_err("fio: connect; %s\n", strerror(errno)); 524 close(fd); 525 return ret; 526 } 527 528 return fd; 529 } 530 531 int fio_client_connect(struct fio_client *client) 532 { 533 int fd; 534 535 dprint(FD_NET, "client: connect to host %s\n", client->hostname); 536 537 if (client->is_sock) 538 fd = fio_client_connect_sock(client); 539 else 540 fd = fio_client_connect_ip(client); 541 542 dprint(FD_NET, "client: %s connected %d\n", client->hostname, fd); 543 544 if (fd < 0) 545 return fd; 546 547 client->fd = fd; 548 fio_client_add_hash(client); 549 client->state = Client_connected; 550 551 probe_client(client); 552 return 0; 553 } 554 555 int fio_client_terminate(struct fio_client *client) 556 { 557 return fio_net_send_quit(client->fd); 558 } 559 560 static void fio_clients_terminate(void) 561 { 562 struct flist_head *entry; 563 struct fio_client *client; 564 565 dprint(FD_NET, "client: terminate clients\n"); 566 567 flist_for_each(entry, &client_list) { 568 client = flist_entry(entry, struct fio_client, list); 569 fio_client_terminate(client); 570 } 571 } 572 573 static void sig_int(int sig) 574 { 575 dprint(FD_NET, "client: got signal %d\n", sig); 576 fio_clients_terminate(); 577 } 578 579 static void client_signal_handler(void) 580 { 581 struct sigaction act; 582 583 memset(&act, 0, sizeof(act)); 584 act.sa_handler = sig_int; 585 act.sa_flags = SA_RESTART; 586 sigaction(SIGINT, &act, NULL); 587 588 memset(&act, 0, sizeof(act)); 589 act.sa_handler = sig_int; 590 act.sa_flags = SA_RESTART; 591 sigaction(SIGTERM, &act, NULL); 592 593 /* Windows uses SIGBREAK as a quit signal from other applications */ 594 #ifdef WIN32 595 memset(&act, 0, sizeof(act)); 596 act.sa_handler = sig_int; 597 act.sa_flags = SA_RESTART; 598 sigaction(SIGBREAK, &act, NULL); 599 #endif 600 601 memset(&act, 0, sizeof(act)); 602 act.sa_handler = sig_show_status; 603 act.sa_flags = SA_RESTART; 604 sigaction(SIGUSR1, &act, NULL); 605 } 606 607 static int send_client_cmd_line(struct fio_client *client) 608 { 609 struct cmd_single_line_pdu *cslp; 610 struct cmd_line_pdu *clp; 611 unsigned long offset; 612 unsigned int *lens; 613 void *pdu; 614 size_t mem; 615 int i, ret; 616 617 dprint(FD_NET, "client: send cmdline %d\n", client->argc); 618 619 lens = malloc(client->argc * sizeof(unsigned int)); 620 621 /* 622 * Find out how much mem we need 623 */ 624 for (i = 0, mem = 0; i < client->argc; i++) { 625 lens[i] = strlen(client->argv[i]) + 1; 626 mem += lens[i]; 627 } 628 629 /* 630 * We need one cmd_line_pdu, and argc number of cmd_single_line_pdu 631 */ 632 mem += sizeof(*clp) + (client->argc * sizeof(*cslp)); 633 634 pdu = malloc(mem); 635 clp = pdu; 636 offset = sizeof(*clp); 637 638 for (i = 0; i < client->argc; i++) { 639 uint16_t arg_len = lens[i]; 640 641 cslp = pdu + offset; 642 strcpy((char *) cslp->text, client->argv[i]); 643 cslp->len = cpu_to_le16(arg_len); 644 offset += sizeof(*cslp) + arg_len; 645 } 646 647 free(lens); 648 clp->lines = cpu_to_le16(client->argc); 649 clp->client_type = __cpu_to_le16(client->type); 650 ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOBLINE, pdu, mem, NULL, NULL); 651 free(pdu); 652 return ret; 653 } 654 655 int fio_clients_connect(void) 656 { 657 struct fio_client *client; 658 struct flist_head *entry, *tmp; 659 int ret; 660 661 #ifdef WIN32 662 WSADATA wsd; 663 WSAStartup(MAKEWORD(2, 2), &wsd); 664 #endif 665 666 dprint(FD_NET, "client: connect all\n"); 667 668 client_signal_handler(); 669 670 flist_for_each_safe(entry, tmp, &client_list) { 671 client = flist_entry(entry, struct fio_client, list); 672 673 ret = fio_client_connect(client); 674 if (ret) { 675 remove_client(client); 676 continue; 677 } 678 679 if (client->argc > 1) 680 send_client_cmd_line(client); 681 } 682 683 return !nr_clients; 684 } 685 686 int fio_start_client(struct fio_client *client) 687 { 688 dprint(FD_NET, "client: start %s\n", client->hostname); 689 return fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_RUN, 0, NULL); 690 } 691 692 int fio_start_all_clients(void) 693 { 694 struct fio_client *client; 695 struct flist_head *entry, *tmp; 696 int ret; 697 698 dprint(FD_NET, "client: start all\n"); 699 700 fio_client_json_init(); 701 702 flist_for_each_safe(entry, tmp, &client_list) { 703 client = flist_entry(entry, struct fio_client, list); 704 705 ret = fio_start_client(client); 706 if (ret) { 707 remove_client(client); 708 continue; 709 } 710 } 711 712 return flist_empty(&client_list); 713 } 714 715 static int __fio_client_send_remote_ini(struct fio_client *client, 716 const char *filename) 717 { 718 struct cmd_load_file_pdu *pdu; 719 size_t p_size; 720 int ret; 721 722 dprint(FD_NET, "send remote ini %s to %s\n", filename, client->hostname); 723 724 p_size = sizeof(*pdu) + strlen(filename) + 1; 725 pdu = malloc(p_size); 726 memset(pdu, 0, p_size); 727 pdu->name_len = strlen(filename); 728 strcpy((char *) pdu->file, filename); 729 pdu->client_type = cpu_to_le16((uint16_t) client->type); 730 731 client->sent_job = 1; 732 ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_LOAD_FILE, pdu, p_size,NULL, NULL); 733 free(pdu); 734 return ret; 735 } 736 737 /* 738 * Send file contents to server backend. We could use sendfile(), but to remain 739 * more portable lets just read/write the darn thing. 740 */ 741 static int __fio_client_send_local_ini(struct fio_client *client, 742 const char *filename) 743 { 744 struct cmd_job_pdu *pdu; 745 size_t p_size; 746 struct stat sb; 747 char *p; 748 void *buf; 749 off_t len; 750 int fd, ret; 751 752 dprint(FD_NET, "send ini %s to %s\n", filename, client->hostname); 753 754 fd = open(filename, O_RDONLY); 755 if (fd < 0) { 756 ret = -errno; 757 log_err("fio: job file <%s> open: %s\n", filename, strerror(errno)); 758 return ret; 759 } 760 761 if (fstat(fd, &sb) < 0) { 762 ret = -errno; 763 log_err("fio: job file stat: %s\n", strerror(errno)); 764 close(fd); 765 return ret; 766 } 767 768 p_size = sb.st_size + sizeof(*pdu); 769 pdu = malloc(p_size); 770 buf = pdu->buf; 771 772 len = sb.st_size; 773 p = buf; 774 if (read_data(fd, p, len)) { 775 log_err("fio: failed reading job file %s\n", filename); 776 close(fd); 777 free(pdu); 778 return 1; 779 } 780 781 pdu->buf_len = __cpu_to_le32(sb.st_size); 782 pdu->client_type = cpu_to_le32(client->type); 783 784 client->sent_job = 1; 785 ret = fio_net_send_cmd(client->fd, FIO_NET_CMD_JOB, pdu, p_size, NULL, NULL); 786 free(pdu); 787 close(fd); 788 return ret; 789 } 790 791 int fio_client_send_ini(struct fio_client *client, const char *filename, 792 bool remote) 793 { 794 int ret; 795 796 if (!remote) 797 ret = __fio_client_send_local_ini(client, filename); 798 else 799 ret = __fio_client_send_remote_ini(client, filename); 800 801 if (!ret) 802 client->sent_job = 1; 803 804 return ret; 805 } 806 807 static int fio_client_send_cf(struct fio_client *client, 808 struct client_file *cf) 809 { 810 return fio_client_send_ini(client, cf->file, cf->remote); 811 } 812 813 int fio_clients_send_ini(const char *filename) 814 { 815 struct fio_client *client; 816 struct flist_head *entry, *tmp; 817 818 flist_for_each_safe(entry, tmp, &client_list) { 819 bool failed = false; 820 821 client = flist_entry(entry, struct fio_client, list); 822 823 if (client->nr_files) { 824 int i; 825 826 for (i = 0; i < client->nr_files; i++) { 827 struct client_file *cf; 828 829 cf = &client->files[i]; 830 831 if (fio_client_send_cf(client, cf)) { 832 failed = true; 833 remove_client(client); 834 break; 835 } 836 } 837 } 838 if (client->sent_job || failed) 839 continue; 840 if (!filename || fio_client_send_ini(client, filename, 0)) 841 remove_client(client); 842 } 843 844 return !nr_clients; 845 } 846 847 int fio_client_update_options(struct fio_client *client, 848 struct thread_options *o, uint64_t *tag) 849 { 850 struct cmd_add_job_pdu pdu; 851 852 pdu.thread_number = cpu_to_le32(client->thread_number); 853 pdu.groupid = cpu_to_le32(client->groupid); 854 convert_thread_options_to_net(&pdu.top, o); 855 856 return fio_net_send_cmd(client->fd, FIO_NET_CMD_UPDATE_JOB, &pdu, sizeof(pdu), tag, &client->cmd_list); 857 } 858 859 static void convert_io_stat(struct io_stat *dst, struct io_stat *src) 860 { 861 dst->max_val = le64_to_cpu(src->max_val); 862 dst->min_val = le64_to_cpu(src->min_val); 863 dst->samples = le64_to_cpu(src->samples); 864 865 /* 866 * Floats arrive as IEEE 754 encoded uint64_t, convert back to double 867 */ 868 dst->mean.u.f = fio_uint64_to_double(le64_to_cpu(dst->mean.u.i)); 869 dst->S.u.f = fio_uint64_to_double(le64_to_cpu(dst->S.u.i)); 870 } 871 872 static void convert_ts(struct thread_stat *dst, struct thread_stat *src) 873 { 874 int i, j; 875 876 dst->error = le32_to_cpu(src->error); 877 dst->thread_number = le32_to_cpu(src->thread_number); 878 dst->groupid = le32_to_cpu(src->groupid); 879 dst->pid = le32_to_cpu(src->pid); 880 dst->members = le32_to_cpu(src->members); 881 dst->unified_rw_rep = le32_to_cpu(src->unified_rw_rep); 882 883 for (i = 0; i < DDIR_RWDIR_CNT; i++) { 884 convert_io_stat(&dst->clat_stat[i], &src->clat_stat[i]); 885 convert_io_stat(&dst->slat_stat[i], &src->slat_stat[i]); 886 convert_io_stat(&dst->lat_stat[i], &src->lat_stat[i]); 887 convert_io_stat(&dst->bw_stat[i], &src->bw_stat[i]); 888 } 889 890 dst->usr_time = le64_to_cpu(src->usr_time); 891 dst->sys_time = le64_to_cpu(src->sys_time); 892 dst->ctx = le64_to_cpu(src->ctx); 893 dst->minf = le64_to_cpu(src->minf); 894 dst->majf = le64_to_cpu(src->majf); 895 dst->clat_percentiles = le64_to_cpu(src->clat_percentiles); 896 dst->percentile_precision = le64_to_cpu(src->percentile_precision); 897 898 for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) { 899 fio_fp64_t *fps = &src->percentile_list[i]; 900 fio_fp64_t *fpd = &dst->percentile_list[i]; 901 902 fpd->u.f = fio_uint64_to_double(le64_to_cpu(fps->u.i)); 903 } 904 905 for (i = 0; i < FIO_IO_U_MAP_NR; i++) { 906 dst->io_u_map[i] = le32_to_cpu(src->io_u_map[i]); 907 dst->io_u_submit[i] = le32_to_cpu(src->io_u_submit[i]); 908 dst->io_u_complete[i] = le32_to_cpu(src->io_u_complete[i]); 909 } 910 911 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) 912 dst->io_u_lat_u[i] = le32_to_cpu(src->io_u_lat_u[i]); 913 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++) 914 dst->io_u_lat_m[i] = le32_to_cpu(src->io_u_lat_m[i]); 915 916 for (i = 0; i < DDIR_RWDIR_CNT; i++) 917 for (j = 0; j < FIO_IO_U_PLAT_NR; j++) 918 dst->io_u_plat[i][j] = le32_to_cpu(src->io_u_plat[i][j]); 919 920 for (i = 0; i < DDIR_RWDIR_CNT; i++) { 921 dst->total_io_u[i] = le64_to_cpu(src->total_io_u[i]); 922 dst->short_io_u[i] = le64_to_cpu(src->short_io_u[i]); 923 dst->drop_io_u[i] = le64_to_cpu(src->drop_io_u[i]); 924 } 925 926 dst->total_submit = le64_to_cpu(src->total_submit); 927 dst->total_complete = le64_to_cpu(src->total_complete); 928 929 for (i = 0; i < DDIR_RWDIR_CNT; i++) { 930 dst->io_bytes[i] = le64_to_cpu(src->io_bytes[i]); 931 dst->runtime[i] = le64_to_cpu(src->runtime[i]); 932 } 933 934 dst->total_run_time = le64_to_cpu(src->total_run_time); 935 dst->continue_on_error = le16_to_cpu(src->continue_on_error); 936 dst->total_err_count = le64_to_cpu(src->total_err_count); 937 dst->first_error = le32_to_cpu(src->first_error); 938 dst->kb_base = le32_to_cpu(src->kb_base); 939 dst->unit_base = le32_to_cpu(src->unit_base); 940 941 dst->latency_depth = le32_to_cpu(src->latency_depth); 942 dst->latency_target = le64_to_cpu(src->latency_target); 943 dst->latency_window = le64_to_cpu(src->latency_window); 944 dst->latency_percentile.u.f = fio_uint64_to_double(le64_to_cpu(src->latency_percentile.u.i)); 945 946 dst->nr_block_infos = le64_to_cpu(src->nr_block_infos); 947 for (i = 0; i < dst->nr_block_infos; i++) 948 dst->block_infos[i] = le32_to_cpu(src->block_infos[i]); 949 950 dst->ss_dur = le64_to_cpu(src->ss_dur); 951 dst->ss_state = le32_to_cpu(src->ss_state); 952 dst->ss_head = le32_to_cpu(src->ss_head); 953 dst->ss_limit.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_limit.u.i)); 954 dst->ss_slope.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_slope.u.i)); 955 dst->ss_deviation.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_deviation.u.i)); 956 dst->ss_criterion.u.f = fio_uint64_to_double(le64_to_cpu(src->ss_criterion.u.i)); 957 958 if (dst->ss_state & __FIO_SS_DATA) { 959 for (i = 0; i < dst->ss_dur; i++ ) { 960 dst->ss_iops_data[i] = le64_to_cpu(src->ss_iops_data[i]); 961 dst->ss_bw_data[i] = le64_to_cpu(src->ss_bw_data[i]); 962 } 963 } 964 } 965 966 static void convert_gs(struct group_run_stats *dst, struct group_run_stats *src) 967 { 968 int i; 969 970 for (i = 0; i < DDIR_RWDIR_CNT; i++) { 971 dst->max_run[i] = le64_to_cpu(src->max_run[i]); 972 dst->min_run[i] = le64_to_cpu(src->min_run[i]); 973 dst->max_bw[i] = le64_to_cpu(src->max_bw[i]); 974 dst->min_bw[i] = le64_to_cpu(src->min_bw[i]); 975 dst->iobytes[i] = le64_to_cpu(src->iobytes[i]); 976 dst->agg[i] = le64_to_cpu(src->agg[i]); 977 } 978 979 dst->kb_base = le32_to_cpu(src->kb_base); 980 dst->unit_base = le32_to_cpu(src->unit_base); 981 dst->groupid = le32_to_cpu(src->groupid); 982 dst->unified_rw_rep = le32_to_cpu(src->unified_rw_rep); 983 } 984 985 static void json_object_add_client_info(struct json_object *obj, 986 struct fio_client *client) 987 { 988 const char *hostname = client->hostname ? client->hostname : ""; 989 990 json_object_add_value_string(obj, "hostname", hostname); 991 json_object_add_value_int(obj, "port", client->port); 992 } 993 994 static void handle_ts(struct fio_client *client, struct fio_net_cmd *cmd) 995 { 996 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload; 997 struct flist_head *opt_list = NULL; 998 struct json_object *tsobj; 999 1000 if (client->opt_lists && p->ts.thread_number <= client->jobs) 1001 opt_list = &client->opt_lists[p->ts.thread_number - 1]; 1002 1003 tsobj = show_thread_status(&p->ts, &p->rs, opt_list, NULL); 1004 client->did_stat = 1; 1005 if (tsobj) { 1006 json_object_add_client_info(tsobj, client); 1007 json_array_add_value_object(clients_array, tsobj); 1008 } 1009 1010 if (sum_stat_clients <= 1) 1011 return; 1012 1013 sum_thread_stats(&client_ts, &p->ts, sum_stat_nr == 1); 1014 sum_group_stats(&client_gs, &p->rs); 1015 1016 client_ts.members++; 1017 client_ts.thread_number = p->ts.thread_number; 1018 client_ts.groupid = p->ts.groupid; 1019 client_ts.unified_rw_rep = p->ts.unified_rw_rep; 1020 1021 if (++sum_stat_nr == sum_stat_clients) { 1022 strcpy(client_ts.name, "All clients"); 1023 tsobj = show_thread_status(&client_ts, &client_gs, NULL, NULL); 1024 if (tsobj) { 1025 json_object_add_client_info(tsobj, client); 1026 json_array_add_value_object(clients_array, tsobj); 1027 } 1028 } 1029 } 1030 1031 static void handle_gs(struct fio_client *client, struct fio_net_cmd *cmd) 1032 { 1033 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload; 1034 1035 if (output_format & FIO_OUTPUT_NORMAL) 1036 show_group_stats(gs, NULL); 1037 } 1038 1039 static void handle_job_opt(struct fio_client *client, struct fio_net_cmd *cmd) 1040 { 1041 struct cmd_job_option *pdu = (struct cmd_job_option *) cmd->payload; 1042 struct print_option *p; 1043 1044 if (!job_opt_object) 1045 return; 1046 1047 pdu->global = le16_to_cpu(pdu->global); 1048 pdu->truncated = le16_to_cpu(pdu->truncated); 1049 pdu->groupid = le32_to_cpu(pdu->groupid); 1050 1051 p = malloc(sizeof(*p)); 1052 p->name = strdup((char *) pdu->name); 1053 if (pdu->value[0] != '\0') 1054 p->value = strdup((char *) pdu->value); 1055 else 1056 p->value = NULL; 1057 1058 if (pdu->global) { 1059 const char *pos = ""; 1060 1061 if (p->value) 1062 pos = p->value; 1063 1064 json_object_add_value_string(job_opt_object, p->name, pos); 1065 } else if (client->opt_lists) { 1066 struct flist_head *opt_list = &client->opt_lists[pdu->groupid]; 1067 1068 flist_add_tail(&p->list, opt_list); 1069 } 1070 } 1071 1072 static void handle_text(struct fio_client *client, struct fio_net_cmd *cmd) 1073 { 1074 struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload; 1075 const char *buf = (const char *) pdu->buf; 1076 const char *name; 1077 int fio_unused ret; 1078 1079 name = client->name ? client->name : client->hostname; 1080 1081 if (!client->skip_newline) 1082 fprintf(f_out, "<%s> ", name); 1083 ret = fwrite(buf, pdu->buf_len, 1, f_out); 1084 fflush(f_out); 1085 client->skip_newline = strchr(buf, '\n') == NULL; 1086 } 1087 1088 static void convert_agg(struct disk_util_agg *agg) 1089 { 1090 int i; 1091 1092 for (i = 0; i < 2; i++) { 1093 agg->ios[i] = le64_to_cpu(agg->ios[i]); 1094 agg->merges[i] = le64_to_cpu(agg->merges[i]); 1095 agg->sectors[i] = le64_to_cpu(agg->sectors[i]); 1096 agg->ticks[i] = le64_to_cpu(agg->ticks[i]); 1097 } 1098 1099 agg->io_ticks = le64_to_cpu(agg->io_ticks); 1100 agg->time_in_queue = le64_to_cpu(agg->time_in_queue); 1101 agg->slavecount = le32_to_cpu(agg->slavecount); 1102 agg->max_util.u.f = fio_uint64_to_double(le64_to_cpu(agg->max_util.u.i)); 1103 } 1104 1105 static void convert_dus(struct disk_util_stat *dus) 1106 { 1107 int i; 1108 1109 for (i = 0; i < 2; i++) { 1110 dus->s.ios[i] = le64_to_cpu(dus->s.ios[i]); 1111 dus->s.merges[i] = le64_to_cpu(dus->s.merges[i]); 1112 dus->s.sectors[i] = le64_to_cpu(dus->s.sectors[i]); 1113 dus->s.ticks[i] = le64_to_cpu(dus->s.ticks[i]); 1114 } 1115 1116 dus->s.io_ticks = le64_to_cpu(dus->s.io_ticks); 1117 dus->s.time_in_queue = le64_to_cpu(dus->s.time_in_queue); 1118 dus->s.msec = le64_to_cpu(dus->s.msec); 1119 } 1120 1121 static void handle_du(struct fio_client *client, struct fio_net_cmd *cmd) 1122 { 1123 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload; 1124 1125 if (!client->disk_stats_shown) { 1126 client->disk_stats_shown = 1; 1127 log_info("\nDisk stats (read/write):\n"); 1128 } 1129 1130 if (output_format & FIO_OUTPUT_JSON) { 1131 struct json_object *duobj; 1132 json_array_add_disk_util(&du->dus, &du->agg, du_array); 1133 duobj = json_array_last_value_object(du_array); 1134 json_object_add_client_info(duobj, client); 1135 } 1136 if (output_format & FIO_OUTPUT_TERSE) 1137 print_disk_util(&du->dus, &du->agg, 1, NULL); 1138 if (output_format & FIO_OUTPUT_NORMAL) 1139 print_disk_util(&du->dus, &du->agg, 0, NULL); 1140 } 1141 1142 static void convert_jobs_eta(struct jobs_eta *je) 1143 { 1144 int i; 1145 1146 je->nr_running = le32_to_cpu(je->nr_running); 1147 je->nr_ramp = le32_to_cpu(je->nr_ramp); 1148 je->nr_pending = le32_to_cpu(je->nr_pending); 1149 je->nr_setting_up = le32_to_cpu(je->nr_setting_up); 1150 je->files_open = le32_to_cpu(je->files_open); 1151 1152 for (i = 0; i < DDIR_RWDIR_CNT; i++) { 1153 je->m_rate[i] = le64_to_cpu(je->m_rate[i]); 1154 je->t_rate[i] = le64_to_cpu(je->t_rate[i]); 1155 je->m_iops[i] = le32_to_cpu(je->m_iops[i]); 1156 je->t_iops[i] = le32_to_cpu(je->t_iops[i]); 1157 je->rate[i] = le64_to_cpu(je->rate[i]); 1158 je->iops[i] = le32_to_cpu(je->iops[i]); 1159 } 1160 1161 je->elapsed_sec = le64_to_cpu(je->elapsed_sec); 1162 je->eta_sec = le64_to_cpu(je->eta_sec); 1163 je->nr_threads = le32_to_cpu(je->nr_threads); 1164 je->is_pow2 = le32_to_cpu(je->is_pow2); 1165 je->unit_base = le32_to_cpu(je->unit_base); 1166 } 1167 1168 void fio_client_sum_jobs_eta(struct jobs_eta *dst, struct jobs_eta *je) 1169 { 1170 int i; 1171 1172 dst->nr_running += je->nr_running; 1173 dst->nr_ramp += je->nr_ramp; 1174 dst->nr_pending += je->nr_pending; 1175 dst->nr_setting_up += je->nr_setting_up; 1176 dst->files_open += je->files_open; 1177 1178 for (i = 0; i < DDIR_RWDIR_CNT; i++) { 1179 dst->m_rate[i] += je->m_rate[i]; 1180 dst->t_rate[i] += je->t_rate[i]; 1181 dst->m_iops[i] += je->m_iops[i]; 1182 dst->t_iops[i] += je->t_iops[i]; 1183 dst->rate[i] += je->rate[i]; 1184 dst->iops[i] += je->iops[i]; 1185 } 1186 1187 dst->elapsed_sec += je->elapsed_sec; 1188 1189 if (je->eta_sec > dst->eta_sec) 1190 dst->eta_sec = je->eta_sec; 1191 1192 dst->nr_threads += je->nr_threads; 1193 1194 /* 1195 * This wont be correct for multiple strings, but at least it 1196 * works for the basic cases. 1197 */ 1198 strcpy((char *) dst->run_str, (char *) je->run_str); 1199 } 1200 1201 static bool remove_reply_cmd(struct fio_client *client, struct fio_net_cmd *cmd) 1202 { 1203 struct fio_net_cmd_reply *reply = NULL; 1204 struct flist_head *entry; 1205 1206 flist_for_each(entry, &client->cmd_list) { 1207 reply = flist_entry(entry, struct fio_net_cmd_reply, list); 1208 1209 if (cmd->tag == (uintptr_t) reply) 1210 break; 1211 1212 reply = NULL; 1213 } 1214 1215 if (!reply) { 1216 log_err("fio: client: unable to find matching tag (%llx)\n", (unsigned long long) cmd->tag); 1217 return false; 1218 } 1219 1220 flist_del(&reply->list); 1221 cmd->tag = reply->saved_tag; 1222 free(reply); 1223 return true; 1224 } 1225 1226 int fio_client_wait_for_reply(struct fio_client *client, uint64_t tag) 1227 { 1228 do { 1229 struct fio_net_cmd_reply *reply = NULL; 1230 struct flist_head *entry; 1231 1232 flist_for_each(entry, &client->cmd_list) { 1233 reply = flist_entry(entry, struct fio_net_cmd_reply, list); 1234 1235 if (tag == (uintptr_t) reply) 1236 break; 1237 1238 reply = NULL; 1239 } 1240 1241 if (!reply) 1242 break; 1243 1244 usleep(1000); 1245 } while (1); 1246 1247 return 0; 1248 } 1249 1250 static void handle_eta(struct fio_client *client, struct fio_net_cmd *cmd) 1251 { 1252 struct jobs_eta *je = (struct jobs_eta *) cmd->payload; 1253 struct client_eta *eta = (struct client_eta *) (uintptr_t) cmd->tag; 1254 1255 dprint(FD_NET, "client: got eta tag %p, %d\n", eta, eta->pending); 1256 1257 assert(client->eta_in_flight == eta); 1258 1259 client->eta_in_flight = NULL; 1260 flist_del_init(&client->eta_list); 1261 client->eta_timeouts = 0; 1262 1263 if (client->ops->jobs_eta) 1264 client->ops->jobs_eta(client, je); 1265 1266 fio_client_sum_jobs_eta(&eta->eta, je); 1267 fio_client_dec_jobs_eta(eta, client->ops->eta); 1268 } 1269 1270 static void client_flush_hist_samples(FILE *f, int hist_coarseness, void *samples, 1271 uint64_t sample_size) 1272 { 1273 struct io_sample *s; 1274 int log_offset; 1275 uint64_t i, j, nr_samples; 1276 struct io_u_plat_entry *entry; 1277 unsigned int *io_u_plat; 1278 1279 int stride = 1 << hist_coarseness; 1280 1281 if (!sample_size) 1282 return; 1283 1284 s = __get_sample(samples, 0, 0); 1285 log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0; 1286 1287 nr_samples = sample_size / __log_entry_sz(log_offset); 1288 1289 for (i = 0; i < nr_samples; i++) { 1290 1291 s = (struct io_sample *)((char *)__get_sample(samples, log_offset, i) + 1292 i * sizeof(struct io_u_plat_entry)); 1293 1294 entry = s->data.plat_entry; 1295 io_u_plat = entry->io_u_plat; 1296 1297 fprintf(f, "%lu, %u, %u, ", (unsigned long) s->time, 1298 io_sample_ddir(s), s->bs); 1299 for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) { 1300 fprintf(f, "%lu, ", hist_sum(j, stride, io_u_plat, NULL)); 1301 } 1302 fprintf(f, "%lu\n", (unsigned long) 1303 hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat, NULL)); 1304 1305 } 1306 } 1307 1308 static int fio_client_handle_iolog(struct fio_client *client, 1309 struct fio_net_cmd *cmd) 1310 { 1311 struct cmd_iolog_pdu *pdu; 1312 bool store_direct; 1313 char *log_pathname; 1314 1315 pdu = convert_iolog(cmd, &store_direct); 1316 if (!pdu) { 1317 log_err("fio: failed converting IO log\n"); 1318 return 1; 1319 } 1320 1321 /* allocate buffer big enough for next sprintf() call */ 1322 log_pathname = malloc(10 + strlen((char *)pdu->name) + 1323 strlen(client->hostname)); 1324 if (!log_pathname) { 1325 log_err("fio: memory allocation of unique pathname failed\n"); 1326 return -1; 1327 } 1328 /* generate a unique pathname for the log file using hostname */ 1329 sprintf(log_pathname, "%s.%s", pdu->name, client->hostname); 1330 1331 if (store_direct) { 1332 ssize_t ret; 1333 size_t sz; 1334 int fd; 1335 1336 fd = open((const char *) log_pathname, 1337 O_WRONLY | O_CREAT | O_TRUNC, 0644); 1338 if (fd < 0) { 1339 log_err("fio: open log %s: %s\n", 1340 log_pathname, strerror(errno)); 1341 return 1; 1342 } 1343 1344 sz = cmd->pdu_len - sizeof(*pdu); 1345 ret = write(fd, pdu->samples, sz); 1346 close(fd); 1347 1348 if (ret != sz) { 1349 log_err("fio: short write on compressed log\n"); 1350 return 1; 1351 } 1352 1353 return 0; 1354 } else { 1355 FILE *f; 1356 f = fopen((const char *) log_pathname, "w"); 1357 if (!f) { 1358 log_err("fio: fopen log %s : %s\n", 1359 log_pathname, strerror(errno)); 1360 return 1; 1361 } 1362 1363 if (pdu->log_type == IO_LOG_TYPE_HIST) { 1364 client_flush_hist_samples(f, pdu->log_hist_coarseness, pdu->samples, 1365 pdu->nr_samples * sizeof(struct io_sample)); 1366 } else { 1367 flush_samples(f, pdu->samples, 1368 pdu->nr_samples * sizeof(struct io_sample)); 1369 } 1370 fclose(f); 1371 return 0; 1372 } 1373 } 1374 1375 static void handle_probe(struct fio_client *client, struct fio_net_cmd *cmd) 1376 { 1377 struct cmd_probe_reply_pdu *probe = (struct cmd_probe_reply_pdu *) cmd->payload; 1378 const char *os, *arch; 1379 char bit[16]; 1380 1381 os = fio_get_os_string(probe->os); 1382 if (!os) 1383 os = "unknown"; 1384 1385 arch = fio_get_arch_string(probe->arch); 1386 if (!arch) 1387 os = "unknown"; 1388 1389 sprintf(bit, "%d-bit", probe->bpp * 8); 1390 probe->flags = le64_to_cpu(probe->flags); 1391 1392 log_info("hostname=%s, be=%u, %s, os=%s, arch=%s, fio=%s, flags=%lx\n", 1393 probe->hostname, probe->bigendian, bit, os, arch, 1394 probe->fio_version, (unsigned long) probe->flags); 1395 1396 if (!client->name) 1397 client->name = strdup((char *) probe->hostname); 1398 } 1399 1400 static void handle_start(struct fio_client *client, struct fio_net_cmd *cmd) 1401 { 1402 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload; 1403 1404 client->state = Client_started; 1405 client->jobs = le32_to_cpu(pdu->jobs); 1406 client->nr_stat = le32_to_cpu(pdu->stat_outputs); 1407 1408 if (client->jobs) { 1409 int i; 1410 1411 if (client->opt_lists) 1412 free(client->opt_lists); 1413 1414 client->opt_lists = malloc(client->jobs * sizeof(struct flist_head)); 1415 for (i = 0; i < client->jobs; i++) 1416 INIT_FLIST_HEAD(&client->opt_lists[i]); 1417 } 1418 1419 sum_stat_clients += client->nr_stat; 1420 } 1421 1422 static void handle_stop(struct fio_client *client, struct fio_net_cmd *cmd) 1423 { 1424 if (client->error) 1425 log_info("client <%s>: exited with error %d\n", client->hostname, client->error); 1426 } 1427 1428 static void convert_stop(struct fio_net_cmd *cmd) 1429 { 1430 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload; 1431 1432 pdu->error = le32_to_cpu(pdu->error); 1433 } 1434 1435 static void convert_text(struct fio_net_cmd *cmd) 1436 { 1437 struct cmd_text_pdu *pdu = (struct cmd_text_pdu *) cmd->payload; 1438 1439 pdu->level = le32_to_cpu(pdu->level); 1440 pdu->buf_len = le32_to_cpu(pdu->buf_len); 1441 pdu->log_sec = le64_to_cpu(pdu->log_sec); 1442 pdu->log_usec = le64_to_cpu(pdu->log_usec); 1443 } 1444 1445 static struct cmd_iolog_pdu *convert_iolog_gz(struct fio_net_cmd *cmd, 1446 struct cmd_iolog_pdu *pdu) 1447 { 1448 #ifdef CONFIG_ZLIB 1449 struct cmd_iolog_pdu *ret; 1450 z_stream stream; 1451 uint32_t nr_samples; 1452 size_t total; 1453 void *p; 1454 1455 stream.zalloc = Z_NULL; 1456 stream.zfree = Z_NULL; 1457 stream.opaque = Z_NULL; 1458 stream.avail_in = 0; 1459 stream.next_in = Z_NULL; 1460 1461 if (inflateInit(&stream) != Z_OK) 1462 return NULL; 1463 1464 /* 1465 * Get header first, it's not compressed 1466 */ 1467 nr_samples = le64_to_cpu(pdu->nr_samples); 1468 1469 if (pdu->log_type == IO_LOG_TYPE_HIST) 1470 total = nr_samples * (__log_entry_sz(le32_to_cpu(pdu->log_offset)) + 1471 sizeof(struct io_u_plat_entry)); 1472 else 1473 total = nr_samples * __log_entry_sz(le32_to_cpu(pdu->log_offset)); 1474 ret = malloc(total + sizeof(*pdu)); 1475 ret->nr_samples = nr_samples; 1476 1477 memcpy(ret, pdu, sizeof(*pdu)); 1478 1479 p = (void *) ret + sizeof(*pdu); 1480 1481 stream.avail_in = cmd->pdu_len - sizeof(*pdu); 1482 stream.next_in = (void *) pdu + sizeof(*pdu); 1483 while (stream.avail_in) { 1484 unsigned int this_chunk = 65536; 1485 unsigned int this_len; 1486 int err; 1487 1488 if (this_chunk > total) 1489 this_chunk = total; 1490 1491 stream.avail_out = this_chunk; 1492 stream.next_out = p; 1493 err = inflate(&stream, Z_NO_FLUSH); 1494 /* may be Z_OK, or Z_STREAM_END */ 1495 if (err < 0) { 1496 log_err("fio: inflate error %d\n", err); 1497 free(ret); 1498 ret = NULL; 1499 goto err; 1500 } 1501 1502 this_len = this_chunk - stream.avail_out; 1503 p += this_len; 1504 total -= this_len; 1505 } 1506 1507 err: 1508 inflateEnd(&stream); 1509 return ret; 1510 #else 1511 return NULL; 1512 #endif 1513 } 1514 1515 /* 1516 * This has been compressed on the server side, since it can be big. 1517 * Uncompress here. 1518 */ 1519 static struct cmd_iolog_pdu *convert_iolog(struct fio_net_cmd *cmd, 1520 bool *store_direct) 1521 { 1522 struct cmd_iolog_pdu *pdu = (struct cmd_iolog_pdu *) cmd->payload; 1523 struct cmd_iolog_pdu *ret; 1524 uint64_t i; 1525 int compressed; 1526 void *samples; 1527 1528 *store_direct = false; 1529 1530 /* 1531 * Convert if compressed and we support it. If it's not 1532 * compressed, we need not do anything. 1533 */ 1534 compressed = le32_to_cpu(pdu->compressed); 1535 if (compressed == XMIT_COMPRESSED) { 1536 #ifndef CONFIG_ZLIB 1537 log_err("fio: server sent compressed data by mistake\n"); 1538 return NULL; 1539 #endif 1540 ret = convert_iolog_gz(cmd, pdu); 1541 if (!ret) { 1542 log_err("fio: failed decompressing log\n"); 1543 return NULL; 1544 } 1545 } else if (compressed == STORE_COMPRESSED) { 1546 *store_direct = true; 1547 ret = pdu; 1548 } else 1549 ret = pdu; 1550 1551 ret->nr_samples = le64_to_cpu(ret->nr_samples); 1552 ret->thread_number = le32_to_cpu(ret->thread_number); 1553 ret->log_type = le32_to_cpu(ret->log_type); 1554 ret->compressed = le32_to_cpu(ret->compressed); 1555 ret->log_offset = le32_to_cpu(ret->log_offset); 1556 ret->log_hist_coarseness = le32_to_cpu(ret->log_hist_coarseness); 1557 1558 if (*store_direct) 1559 return ret; 1560 1561 samples = &ret->samples[0]; 1562 for (i = 0; i < ret->nr_samples; i++) { 1563 struct io_sample *s; 1564 1565 s = __get_sample(samples, ret->log_offset, i); 1566 if (ret->log_type == IO_LOG_TYPE_HIST) 1567 s = (struct io_sample *)((void *)s + sizeof(struct io_u_plat_entry) * i); 1568 1569 s->time = le64_to_cpu(s->time); 1570 s->data.val = le64_to_cpu(s->data.val); 1571 s->__ddir = le32_to_cpu(s->__ddir); 1572 s->bs = le32_to_cpu(s->bs); 1573 1574 if (ret->log_offset) { 1575 struct io_sample_offset *so = (void *) s; 1576 1577 so->offset = le64_to_cpu(so->offset); 1578 } 1579 1580 if (ret->log_type == IO_LOG_TYPE_HIST) { 1581 s->data.plat_entry = (struct io_u_plat_entry *)(((void *)s) + sizeof(*s)); 1582 s->data.plat_entry->list.next = NULL; 1583 s->data.plat_entry->list.prev = NULL; 1584 } 1585 } 1586 1587 return ret; 1588 } 1589 1590 static void sendfile_reply(int fd, struct cmd_sendfile_reply *rep, 1591 size_t size, uint64_t tag) 1592 { 1593 rep->error = cpu_to_le32(rep->error); 1594 fio_net_send_cmd(fd, FIO_NET_CMD_SENDFILE, rep, size, &tag, NULL); 1595 } 1596 1597 static int fio_send_file(struct fio_client *client, struct cmd_sendfile *pdu, 1598 uint64_t tag) 1599 { 1600 struct cmd_sendfile_reply *rep; 1601 struct stat sb; 1602 size_t size; 1603 int fd; 1604 1605 size = sizeof(*rep); 1606 rep = malloc(size); 1607 1608 if (stat((char *)pdu->path, &sb) < 0) { 1609 fail: 1610 rep->error = errno; 1611 sendfile_reply(client->fd, rep, size, tag); 1612 free(rep); 1613 return 1; 1614 } 1615 1616 size += sb.st_size; 1617 rep = realloc(rep, size); 1618 rep->size = cpu_to_le32((uint32_t) sb.st_size); 1619 1620 fd = open((char *)pdu->path, O_RDONLY); 1621 if (fd == -1 ) 1622 goto fail; 1623 1624 rep->error = read_data(fd, &rep->data, sb.st_size); 1625 sendfile_reply(client->fd, rep, size, tag); 1626 free(rep); 1627 close(fd); 1628 return 0; 1629 } 1630 1631 int fio_handle_client(struct fio_client *client) 1632 { 1633 struct client_ops *ops = client->ops; 1634 struct fio_net_cmd *cmd; 1635 int size; 1636 1637 dprint(FD_NET, "client: handle %s\n", client->hostname); 1638 1639 cmd = fio_net_recv_cmd(client->fd, true); 1640 if (!cmd) 1641 return 0; 1642 1643 dprint(FD_NET, "client: got cmd op %s from %s (pdu=%u)\n", 1644 fio_server_op(cmd->opcode), client->hostname, cmd->pdu_len); 1645 1646 switch (cmd->opcode) { 1647 case FIO_NET_CMD_QUIT: 1648 if (ops->quit) 1649 ops->quit(client, cmd); 1650 remove_client(client); 1651 break; 1652 case FIO_NET_CMD_TEXT: 1653 convert_text(cmd); 1654 ops->text(client, cmd); 1655 break; 1656 case FIO_NET_CMD_DU: { 1657 struct cmd_du_pdu *du = (struct cmd_du_pdu *) cmd->payload; 1658 1659 convert_dus(&du->dus); 1660 convert_agg(&du->agg); 1661 1662 ops->disk_util(client, cmd); 1663 break; 1664 } 1665 case FIO_NET_CMD_TS: { 1666 struct cmd_ts_pdu *p = (struct cmd_ts_pdu *) cmd->payload; 1667 1668 dprint(FD_NET, "client: ts->ss_state = %u\n", (unsigned int) le32_to_cpu(p->ts.ss_state)); 1669 if (le32_to_cpu(p->ts.ss_state) & __FIO_SS_DATA) { 1670 dprint(FD_NET, "client: received steadystate ring buffers\n"); 1671 1672 size = le64_to_cpu(p->ts.ss_dur); 1673 p->ts.ss_iops_data = (uint64_t *) ((struct cmd_ts_pdu *)cmd->payload + 1); 1674 p->ts.ss_bw_data = p->ts.ss_iops_data + size; 1675 } 1676 1677 convert_ts(&p->ts, &p->ts); 1678 convert_gs(&p->rs, &p->rs); 1679 1680 ops->thread_status(client, cmd); 1681 break; 1682 } 1683 case FIO_NET_CMD_GS: { 1684 struct group_run_stats *gs = (struct group_run_stats *) cmd->payload; 1685 1686 convert_gs(gs, gs); 1687 1688 ops->group_stats(client, cmd); 1689 break; 1690 } 1691 case FIO_NET_CMD_ETA: { 1692 struct jobs_eta *je = (struct jobs_eta *) cmd->payload; 1693 1694 if (!remove_reply_cmd(client, cmd)) 1695 break; 1696 convert_jobs_eta(je); 1697 handle_eta(client, cmd); 1698 break; 1699 } 1700 case FIO_NET_CMD_PROBE: 1701 remove_reply_cmd(client, cmd); 1702 ops->probe(client, cmd); 1703 break; 1704 case FIO_NET_CMD_SERVER_START: 1705 client->state = Client_running; 1706 if (ops->job_start) 1707 ops->job_start(client, cmd); 1708 break; 1709 case FIO_NET_CMD_START: { 1710 struct cmd_start_pdu *pdu = (struct cmd_start_pdu *) cmd->payload; 1711 1712 pdu->jobs = le32_to_cpu(pdu->jobs); 1713 ops->start(client, cmd); 1714 break; 1715 } 1716 case FIO_NET_CMD_STOP: { 1717 struct cmd_end_pdu *pdu = (struct cmd_end_pdu *) cmd->payload; 1718 1719 convert_stop(cmd); 1720 client->state = Client_stopped; 1721 client->error = le32_to_cpu(pdu->error); 1722 client->signal = le32_to_cpu(pdu->signal); 1723 ops->stop(client, cmd); 1724 break; 1725 } 1726 case FIO_NET_CMD_ADD_JOB: { 1727 struct cmd_add_job_pdu *pdu = (struct cmd_add_job_pdu *) cmd->payload; 1728 1729 client->thread_number = le32_to_cpu(pdu->thread_number); 1730 client->groupid = le32_to_cpu(pdu->groupid); 1731 1732 if (ops->add_job) 1733 ops->add_job(client, cmd); 1734 break; 1735 } 1736 case FIO_NET_CMD_IOLOG: 1737 fio_client_handle_iolog(client, cmd); 1738 break; 1739 case FIO_NET_CMD_UPDATE_JOB: 1740 ops->update_job(client, cmd); 1741 remove_reply_cmd(client, cmd); 1742 break; 1743 case FIO_NET_CMD_VTRIGGER: { 1744 struct all_io_list *pdu = (struct all_io_list *) cmd->payload; 1745 char buf[128]; 1746 int off = 0; 1747 1748 if (aux_path) { 1749 strcpy(buf, aux_path); 1750 off = strlen(buf); 1751 } 1752 1753 __verify_save_state(pdu, server_name(client, &buf[off], sizeof(buf) - off)); 1754 exec_trigger(trigger_cmd); 1755 break; 1756 } 1757 case FIO_NET_CMD_SENDFILE: { 1758 struct cmd_sendfile *pdu = (struct cmd_sendfile *) cmd->payload; 1759 fio_send_file(client, pdu, cmd->tag); 1760 break; 1761 } 1762 case FIO_NET_CMD_JOB_OPT: { 1763 handle_job_opt(client, cmd); 1764 break; 1765 } 1766 default: 1767 log_err("fio: unknown client op: %s\n", fio_server_op(cmd->opcode)); 1768 break; 1769 } 1770 1771 free(cmd); 1772 return 1; 1773 } 1774 1775 int fio_clients_send_trigger(const char *cmd) 1776 { 1777 struct flist_head *entry; 1778 struct fio_client *client; 1779 size_t slen; 1780 1781 dprint(FD_NET, "client: send vtrigger: %s\n", cmd); 1782 1783 if (!cmd) 1784 slen = 0; 1785 else 1786 slen = strlen(cmd); 1787 1788 flist_for_each(entry, &client_list) { 1789 struct cmd_vtrigger_pdu *pdu; 1790 1791 client = flist_entry(entry, struct fio_client, list); 1792 1793 pdu = malloc(sizeof(*pdu) + slen); 1794 pdu->len = cpu_to_le16((uint16_t) slen); 1795 if (slen) 1796 memcpy(pdu->cmd, cmd, slen); 1797 fio_net_send_cmd(client->fd, FIO_NET_CMD_VTRIGGER, pdu, 1798 sizeof(*pdu) + slen, NULL, NULL); 1799 free(pdu); 1800 } 1801 1802 return 0; 1803 } 1804 1805 static void request_client_etas(struct client_ops *ops) 1806 { 1807 struct fio_client *client; 1808 struct flist_head *entry; 1809 struct client_eta *eta; 1810 int skipped = 0; 1811 1812 dprint(FD_NET, "client: request eta (%d)\n", nr_clients); 1813 1814 eta = calloc(1, sizeof(*eta) + __THREAD_RUNSTR_SZ(REAL_MAX_JOBS)); 1815 eta->pending = nr_clients; 1816 1817 flist_for_each(entry, &client_list) { 1818 client = flist_entry(entry, struct fio_client, list); 1819 1820 if (!flist_empty(&client->eta_list)) { 1821 skipped++; 1822 continue; 1823 } 1824 if (client->state != Client_running) 1825 continue; 1826 1827 assert(!client->eta_in_flight); 1828 flist_add_tail(&client->eta_list, &eta_list); 1829 client->eta_in_flight = eta; 1830 fio_net_send_simple_cmd(client->fd, FIO_NET_CMD_SEND_ETA, 1831 (uintptr_t) eta, &client->cmd_list); 1832 } 1833 1834 while (skipped--) { 1835 if (!fio_client_dec_jobs_eta(eta, ops->eta)) 1836 break; 1837 } 1838 1839 dprint(FD_NET, "client: requested eta tag %p\n", eta); 1840 } 1841 1842 /* 1843 * A single SEND_ETA timeout isn't fatal. Attempt to recover. 1844 */ 1845 static int handle_cmd_timeout(struct fio_client *client, 1846 struct fio_net_cmd_reply *reply) 1847 { 1848 flist_del(&reply->list); 1849 free(reply); 1850 1851 if (reply->opcode != FIO_NET_CMD_SEND_ETA) 1852 return 1; 1853 1854 log_info("client <%s>: timeout on SEND_ETA\n", client->hostname); 1855 1856 flist_del_init(&client->eta_list); 1857 if (client->eta_in_flight) { 1858 fio_client_dec_jobs_eta(client->eta_in_flight, client->ops->eta); 1859 client->eta_in_flight = NULL; 1860 } 1861 1862 /* 1863 * If we fail 5 in a row, give up... 1864 */ 1865 if (client->eta_timeouts++ > 5) 1866 return 1; 1867 1868 return 0; 1869 } 1870 1871 static int client_check_cmd_timeout(struct fio_client *client, 1872 struct timeval *now) 1873 { 1874 struct fio_net_cmd_reply *reply; 1875 struct flist_head *entry, *tmp; 1876 int ret = 0; 1877 1878 flist_for_each_safe(entry, tmp, &client->cmd_list) { 1879 reply = flist_entry(entry, struct fio_net_cmd_reply, list); 1880 1881 if (mtime_since(&reply->tv, now) < FIO_NET_CLIENT_TIMEOUT) 1882 continue; 1883 1884 if (!handle_cmd_timeout(client, reply)) 1885 continue; 1886 1887 log_err("fio: client %s, timeout on cmd %s\n", client->hostname, 1888 fio_server_op(reply->opcode)); 1889 ret = 1; 1890 } 1891 1892 return flist_empty(&client->cmd_list) && ret; 1893 } 1894 1895 static int fio_check_clients_timed_out(void) 1896 { 1897 struct fio_client *client; 1898 struct flist_head *entry, *tmp; 1899 struct timeval tv; 1900 int ret = 0; 1901 1902 fio_gettime(&tv, NULL); 1903 1904 flist_for_each_safe(entry, tmp, &client_list) { 1905 client = flist_entry(entry, struct fio_client, list); 1906 1907 if (flist_empty(&client->cmd_list)) 1908 continue; 1909 1910 if (!client_check_cmd_timeout(client, &tv)) 1911 continue; 1912 1913 if (client->ops->timed_out) 1914 client->ops->timed_out(client); 1915 else 1916 log_err("fio: client %s timed out\n", client->hostname); 1917 1918 client->error = ETIMEDOUT; 1919 remove_client(client); 1920 ret = 1; 1921 } 1922 1923 return ret; 1924 } 1925 1926 int fio_handle_clients(struct client_ops *ops) 1927 { 1928 struct pollfd *pfds; 1929 int i, ret = 0, retval = 0; 1930 1931 fio_gettime(&eta_tv, NULL); 1932 1933 pfds = malloc(nr_clients * sizeof(struct pollfd)); 1934 1935 init_thread_stat(&client_ts); 1936 init_group_run_stat(&client_gs); 1937 1938 while (!exit_backend && nr_clients) { 1939 struct flist_head *entry, *tmp; 1940 struct fio_client *client; 1941 1942 i = 0; 1943 flist_for_each_safe(entry, tmp, &client_list) { 1944 client = flist_entry(entry, struct fio_client, list); 1945 1946 if (!client->sent_job && !client->ops->stay_connected && 1947 flist_empty(&client->cmd_list)) { 1948 remove_client(client); 1949 continue; 1950 } 1951 1952 pfds[i].fd = client->fd; 1953 pfds[i].events = POLLIN; 1954 i++; 1955 } 1956 1957 if (!nr_clients) 1958 break; 1959 1960 assert(i == nr_clients); 1961 1962 do { 1963 struct timeval tv; 1964 int timeout; 1965 1966 fio_gettime(&tv, NULL); 1967 if (mtime_since(&eta_tv, &tv) >= 900) { 1968 request_client_etas(ops); 1969 memcpy(&eta_tv, &tv, sizeof(tv)); 1970 1971 if (fio_check_clients_timed_out()) 1972 break; 1973 } 1974 1975 check_trigger_file(); 1976 1977 timeout = min(100u, ops->eta_msec); 1978 1979 ret = poll(pfds, nr_clients, timeout); 1980 if (ret < 0) { 1981 if (errno == EINTR) 1982 continue; 1983 log_err("fio: poll clients: %s\n", strerror(errno)); 1984 break; 1985 } else if (!ret) 1986 continue; 1987 } while (ret <= 0); 1988 1989 for (i = 0; i < nr_clients; i++) { 1990 if (!(pfds[i].revents & POLLIN)) 1991 continue; 1992 1993 client = find_client_by_fd(pfds[i].fd); 1994 if (!client) { 1995 log_err("fio: unknown client fd %ld\n", (long) pfds[i].fd); 1996 continue; 1997 } 1998 if (!fio_handle_client(client)) { 1999 log_info("client: host=%s disconnected\n", 2000 client->hostname); 2001 remove_client(client); 2002 retval = 1; 2003 } else if (client->error) 2004 retval = 1; 2005 fio_put_client(client); 2006 } 2007 } 2008 2009 fio_client_json_fini(); 2010 2011 free(pfds); 2012 return retval || error_clients; 2013 } 2014