backend.c

   1 /*
   2  * fio - the flexible io tester
   3  *
   4  * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
   5  * Copyright (C) 2006-2012 Jens Axboe <axboe@kernel.dk>
   6  *
   7  * The license below covers all files distributed with fio unless otherwise
   8  * noted in the file itself.
   9  *
  10  *  This program is free software; you can redistribute it and/or modify
  11  *  it under the terms of the GNU General Public License version 2 as
  12  *  published by the Free Software Foundation.
  13  *
  14  *  This program is distributed in the hope that it will be useful,
  15  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *  GNU General Public License for more details.
  18  *
  19  *  You should have received a copy of the GNU General Public License
  20  *  along with this program; if not, write to the Free Software
  21  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  22  *
  23  */
  24 #include <unistd.h>
  25 #include <fcntl.h>
  26 #include <string.h>
  27 #include <limits.h>
  28 #include <signal.h>
  29 #include <time.h>
  30 #include <locale.h>
  31 #include <assert.h>
  32 #include <time.h>
  33 #include <inttypes.h>
  34 #include <sys/stat.h>
  35 #include <sys/wait.h>
  36 #include <sys/ipc.h>
  37 #include <sys/mman.h>
  38
  39 #include "fio.h"
  40 #ifndef FIO_NO_HAVE_SHM_H
  41 #include <sys/shm.h>
  42 #endif
  43 #include "hash.h"
  44 #include "smalloc.h"
  45 #include "verify.h"
  46 #include "trim.h"
  47 #include "diskutil.h"
  48 #include "cgroup.h"
  49 #include "profile.h"
  50 #include "lib/rand.h"
  51 #include "memalign.h"
  52 #include "server.h"
  53 #include "lib/getrusage.h"
  54 #include "idletime.h"
  55 #include "err.h"
  56 #include "lib/tp.h"
  57
  58 static pthread_t helper_thread;
  59 static pthread_mutex_t helper_lock;
  60 pthread_cond_t helper_cond;
  61 int helper_do_stat = 0;
  62
  63 static struct fio_mutex *startup_mutex;
  64 static struct flist_head *cgroup_list;
  65 static char *cgroup_mnt;
  66 static int exit_value;
  67 static volatile int fio_abort;
  68 static unsigned int nr_process = 0;
  69 static unsigned int nr_thread = 0;
  70
  71 struct io_log *agg_io_log[DDIR_RWDIR_CNT];
  72
  73 int groupid = 0;
  74 unsigned int thread_number = 0;
  75 unsigned int stat_number = 0;
  76 int shm_id = 0;
  77 int temp_stall_ts;
  78 unsigned long done_secs = 0;
  79 volatile int helper_exit = 0;
  80
  81 #define PAGE_ALIGN(buf) \
  82         (char *) (((uintptr_t) (buf) + page_mask) & ~page_mask)
  83
  84 #define JOB_START_TIMEOUT       (5 * 1000)
  85
  86 static void sig_int(int sig)
  87 {
  88         if (threads) {
  89                 if (is_backend)
  90                         fio_server_got_signal(sig);
  91                 else {
  92                         log_info("\nfio: terminating on signal %d\n", sig);
  93                         log_info_flush();
  94                         exit_value = 128;
  95                 }
  96
  97                 fio_terminate_threads(TERMINATE_ALL);
  98         }
  99 }
 100
 101 static void sig_show_status(int sig)
 102 {
 103         show_running_run_stats();
 104 }
 105
 106 static void set_sig_handlers(void)
 107 {
 108         struct sigaction act;
 109
 110         memset(&act, 0, sizeof(act));
 111         act.sa_handler = sig_int;
 112         act.sa_flags = SA_RESTART;
 113         sigaction(SIGINT, &act, NULL);
 114
 115         memset(&act, 0, sizeof(act));
 116         act.sa_handler = sig_int;
 117         act.sa_flags = SA_RESTART;
 118         sigaction(SIGTERM, &act, NULL);
 119
 120 /* Windows uses SIGBREAK as a quit signal from other applications */
 121 #ifdef WIN32
 122         memset(&act, 0, sizeof(act));
 123         act.sa_handler = sig_int;
 124         act.sa_flags = SA_RESTART;
 125         sigaction(SIGBREAK, &act, NULL);
 126 #endif
 127
 128         memset(&act, 0, sizeof(act));
 129         act.sa_handler = sig_show_status;
 130         act.sa_flags = SA_RESTART;
 131         sigaction(SIGUSR1, &act, NULL);
 132
 133         if (is_backend) {
 134                 memset(&act, 0, sizeof(act));
 135                 act.sa_handler = sig_int;
 136                 act.sa_flags = SA_RESTART;
 137                 sigaction(SIGPIPE, &act, NULL);
 138         }
 139 }
 140
 141 /*
 142  * Check if we are above the minimum rate given.
 143  */
 144 static int __check_min_rate(struct thread_data *td, struct timeval *now,
 145                             enum fio_ddir ddir)
 146 {
 147         unsigned long long bytes = 0;
 148         unsigned long iops = 0;
 149         unsigned long spent;
 150         unsigned long rate;
 151         unsigned int ratemin = 0;
 152         unsigned int rate_iops = 0;
 153         unsigned int rate_iops_min = 0;
 154
 155         assert(ddir_rw(ddir));
 156
 157         if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir])
 158                 return 0;
 159
 160         /*
 161          * allow a 2 second settle period in the beginning
 162          */
 163         if (mtime_since(&td->start, now) < 2000)
 164                 return 0;
 165
 166         iops += td->this_io_blocks[ddir];
 167         bytes += td->this_io_bytes[ddir];
 168         ratemin += td->o.ratemin[ddir];
 169         rate_iops += td->o.rate_iops[ddir];
 170         rate_iops_min += td->o.rate_iops_min[ddir];
 171
 172         /*
 173          * if rate blocks is set, sample is running
 174          */
 175         if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) {
 176                 spent = mtime_since(&td->lastrate[ddir], now);
 177                 if (spent < td->o.ratecycle)
 178                         return 0;
 179
 180                 if (td->o.rate[ddir]) {
 181                         /*
 182                          * check bandwidth specified rate
 183                          */
 184                         if (bytes < td->rate_bytes[ddir]) {
 185                                 log_err("%s: min rate %u not met\n", td->o.name,
 186                                                                 ratemin);
 187                                 return 1;
 188                         } else {
 189                                 if (spent)
 190                                         rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent;
 191                                 else
 192                                         rate = 0;
 193
 194                                 if (rate < ratemin ||
 195                                     bytes < td->rate_bytes[ddir]) {
 196                                         log_err("%s: min rate %u not met, got"
 197                                                 " %luKB/sec\n", td->o.name,
 198                                                         ratemin, rate);
 199                                         return 1;
 200                                 }
 201                         }
 202                 } else {
 203                         /*
 204                          * checks iops specified rate
 205                          */
 206                         if (iops < rate_iops) {
 207                                 log_err("%s: min iops rate %u not met\n",
 208                                                 td->o.name, rate_iops);
 209                                 return 1;
 210                         } else {
 211                                 if (spent)
 212                                         rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent;
 213                                 else
 214                                         rate = 0;
 215
 216                                 if (rate < rate_iops_min ||
 217                                     iops < td->rate_blocks[ddir]) {
 218                                         log_err("%s: min iops rate %u not met,"
 219                                                 " got %lu\n", td->o.name,
 220                                                         rate_iops_min, rate);
 221                                 }
 222                         }
 223                 }
 224         }
 225
 226         td->rate_bytes[ddir] = bytes;
 227         td->rate_blocks[ddir] = iops;
 228         memcpy(&td->lastrate[ddir], now, sizeof(*now));
 229         return 0;
 230 }
 231
 232 static int check_min_rate(struct thread_data *td, struct timeval *now,
 233                           uint64_t *bytes_done)
 234 {
 235         int ret = 0;
 236
 237         if (bytes_done[DDIR_READ])
 238                 ret |= __check_min_rate(td, now, DDIR_READ);
 239         if (bytes_done[DDIR_WRITE])
 240                 ret |= __check_min_rate(td, now, DDIR_WRITE);
 241         if (bytes_done[DDIR_TRIM])
 242                 ret |= __check_min_rate(td, now, DDIR_TRIM);
 243
 244         return ret;
 245 }
 246
 247 /*
 248  * When job exits, we can cancel the in-flight IO if we are using async
 249  * io. Attempt to do so.
 250  */
 251 static void cleanup_pending_aio(struct thread_data *td)
 252 {
 253         int r;
 254
 255         /*
 256          * get immediately available events, if any
 257          */
 258         r = io_u_queued_complete(td, 0, NULL);
 259         if (r < 0)
 260                 return;
 261
 262         /*
 263          * now cancel remaining active events
 264          */
 265         if (td->io_ops->cancel) {
 266                 struct io_u *io_u;
 267                 int i;
 268
 269                 io_u_qiter(&td->io_u_all, io_u, i) {
 270                         if (io_u->flags & IO_U_F_FLIGHT) {
 271                                 r = td->io_ops->cancel(td, io_u);
 272                                 if (!r)
 273                                         put_io_u(td, io_u);
 274                         }
 275                 }
 276         }
 277
 278         if (td->cur_depth)
 279                 r = io_u_queued_complete(td, td->cur_depth, NULL);
 280 }
 281
 282 /*
 283  * Helper to handle the final sync of a file. Works just like the normal
 284  * io path, just does everything sync.
 285  */
 286 static int fio_io_sync(struct thread_data *td, struct fio_file *f)
 287 {
 288         struct io_u *io_u = __get_io_u(td);
 289         int ret;
 290
 291         if (!io_u)
 292                 return 1;
 293
 294         io_u->ddir = DDIR_SYNC;
 295         io_u->file = f;
 296
 297         if (td_io_prep(td, io_u)) {
 298                 put_io_u(td, io_u);
 299                 return 1;
 300         }
 301
 302 requeue:
 303         ret = td_io_queue(td, io_u);
 304         if (ret < 0) {
 305                 td_verror(td, io_u->error, "td_io_queue");
 306                 put_io_u(td, io_u);
 307                 return 1;
 308         } else if (ret == FIO_Q_QUEUED) {
 309                 if (io_u_queued_complete(td, 1, NULL) < 0)
 310                         return 1;
 311         } else if (ret == FIO_Q_COMPLETED) {
 312                 if (io_u->error) {
 313                         td_verror(td, io_u->error, "td_io_queue");
 314                         return 1;
 315                 }
 316
 317                 if (io_u_sync_complete(td, io_u, NULL) < 0)
 318                         return 1;
 319         } else if (ret == FIO_Q_BUSY) {
 320                 if (td_io_commit(td))
 321                         return 1;
 322                 goto requeue;
 323         }
 324
 325         return 0;
 326 }
 327
 328 static int fio_file_fsync(struct thread_data *td, struct fio_file *f)
 329 {
 330         int ret;
 331
 332         if (fio_file_open(f))
 333                 return fio_io_sync(td, f);
 334
 335         if (td_io_open_file(td, f))
 336                 return 1;
 337
 338         ret = fio_io_sync(td, f);
 339         td_io_close_file(td, f);
 340         return ret;
 341 }
 342
 343 static inline void __update_tv_cache(struct thread_data *td)
 344 {
 345         fio_gettime(&td->tv_cache, NULL);
 346 }
 347
 348 static inline void update_tv_cache(struct thread_data *td)
 349 {
 350         if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask)
 351                 __update_tv_cache(td);
 352 }
 353
 354 static inline int runtime_exceeded(struct thread_data *td, struct timeval *t)
 355 {
 356         if (in_ramp_time(td))
 357                 return 0;
 358         if (!td->o.timeout)
 359                 return 0;
 360         if (utime_since(&td->epoch, t) >= td->o.timeout)
 361                 return 1;
 362
 363         return 0;
 364 }
 365
 366 static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
 367                                int *retptr)
 368 {
 369         int ret = *retptr;
 370
 371         if (ret < 0 || td->error) {
 372                 int err = td->error;
 373                 enum error_type_bit eb;
 374
 375                 if (ret < 0)
 376                         err = -ret;
 377
 378                 eb = td_error_type(ddir, err);
 379                 if (!(td->o.continue_on_error & (1 << eb)))
 380                         return 1;
 381
 382                 if (td_non_fatal_error(td, eb, err)) {
 383                         /*
 384                          * Continue with the I/Os in case of
 385                          * a non fatal error.
 386                          */
 387                         update_error_count(td, err);
 388                         td_clear_error(td);
 389                         *retptr = 0;
 390                         return 0;
 391                 } else if (td->o.fill_device && err == ENOSPC) {
 392                         /*
 393                          * We expect to hit this error if
 394                          * fill_device option is set.
 395                          */
 396                         td_clear_error(td);
 397                         fio_mark_td_terminate(td);
 398                         return 1;
 399                 } else {
 400                         /*
 401                          * Stop the I/O in case of a fatal
 402                          * error.
 403                          */
 404                         update_error_count(td, err);
 405                         return 1;
 406                 }
 407         }
 408
 409         return 0;
 410 }
 411
 412 static void check_update_rusage(struct thread_data *td)
 413 {
 414         if (td->update_rusage) {
 415                 td->update_rusage = 0;
 416                 update_rusage_stat(td);
 417                 fio_mutex_up(td->rusage_sem);
 418         }
 419 }
 420
 421 /*
 422  * The main verify engine. Runs over the writes we previously submitted,
 423  * reads the blocks back in, and checks the crc/md5 of the data.
 424  */
 425 static void do_verify(struct thread_data *td, uint64_t verify_bytes)
 426 {
 427         uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 };
 428         struct fio_file *f;
 429         struct io_u *io_u;
 430         int ret, min_events;
 431         unsigned int i;
 432
 433         dprint(FD_VERIFY, "starting loop\n");
 434
 435         /*
 436          * sync io first and invalidate cache, to make sure we really
 437          * read from disk.
 438          */
 439         for_each_file(td, f, i) {
 440                 if (!fio_file_open(f))
 441                         continue;
 442                 if (fio_io_sync(td, f))
 443                         break;
 444                 if (file_invalidate_cache(td, f))
 445                         break;
 446         }
 447
 448         check_update_rusage(td);
 449
 450         if (td->error)
 451                 return;
 452
 453         td_set_runstate(td, TD_VERIFYING);
 454
 455         io_u = NULL;
 456         while (!td->terminate) {
 457                 enum fio_ddir ddir;
 458                 int ret2, full;
 459
 460                 update_tv_cache(td);
 461                 check_update_rusage(td);
 462
 463                 if (runtime_exceeded(td, &td->tv_cache)) {
 464                         __update_tv_cache(td);
 465                         if (runtime_exceeded(td, &td->tv_cache)) {
 466                                 fio_mark_td_terminate(td);
 467                                 break;
 468                         }
 469                 }
 470
 471                 if (flow_threshold_exceeded(td))
 472                         continue;
 473
 474                 if (!td->o.experimental_verify) {
 475                         io_u = __get_io_u(td);
 476                         if (!io_u)
 477                                 break;
 478
 479                         if (get_next_verify(td, io_u)) {
 480                                 put_io_u(td, io_u);
 481                                 break;
 482                         }
 483
 484                         if (td_io_prep(td, io_u)) {
 485                                 put_io_u(td, io_u);
 486                                 break;
 487                         }
 488                 } else {
 489                         if (ddir_rw_sum(bytes_done) + td->o.rw_min_bs > verify_bytes)
 490                                 break;
 491
 492                         while ((io_u = get_io_u(td)) != NULL) {
 493                                 if (IS_ERR(io_u)) {
 494                                         io_u = NULL;
 495                                         ret = FIO_Q_BUSY;
 496                                         goto reap;
 497                                 }
 498
 499                                 /*
 500                                  * We are only interested in the places where
 501                                  * we wrote or trimmed IOs. Turn those into
 502                                  * reads for verification purposes.
 503                                  */
 504                                 if (io_u->ddir == DDIR_READ) {
 505                                         /*
 506                                          * Pretend we issued it for rwmix
 507                                          * accounting
 508                                          */
 509                                         td->io_issues[DDIR_READ]++;
 510                                         put_io_u(td, io_u);
 511                                         continue;
 512                                 } else if (io_u->ddir == DDIR_TRIM) {
 513                                         io_u->ddir = DDIR_READ;
 514                                         io_u->flags |= IO_U_F_TRIMMED;
 515                                         break;
 516                                 } else if (io_u->ddir == DDIR_WRITE) {
 517                                         io_u->ddir = DDIR_READ;
 518                                         break;
 519                                 } else {
 520                                         put_io_u(td, io_u);
 521                                         continue;
 522                                 }
 523                         }
 524
 525                         if (!io_u)
 526                                 break;
 527                 }
 528
 529                 if (verify_state_should_stop(td, io_u)) {
 530                         put_io_u(td, io_u);
 531                         break;
 532                 }
 533
 534                 if (td->o.verify_async)
 535                         io_u->end_io = verify_io_u_async;
 536                 else
 537                         io_u->end_io = verify_io_u;
 538
 539                 ddir = io_u->ddir;
 540
 541                 ret = td_io_queue(td, io_u);
 542                 switch (ret) {
 543                 case FIO_Q_COMPLETED:
 544                         if (io_u->error) {
 545                                 ret = -io_u->error;
 546                                 clear_io_u(td, io_u);
 547                         } else if (io_u->resid) {
 548                                 int bytes = io_u->xfer_buflen - io_u->resid;
 549
 550                                 /*
 551                                  * zero read, fail
 552                                  */
 553                                 if (!bytes) {
 554                                         td_verror(td, EIO, "full resid");
 555                                         put_io_u(td, io_u);
 556                                         break;
 557                                 }
 558
 559                                 io_u->xfer_buflen = io_u->resid;
 560                                 io_u->xfer_buf += bytes;
 561                                 io_u->offset += bytes;
 562
 563                                 if (ddir_rw(io_u->ddir))
 564                                         td->ts.short_io_u[io_u->ddir]++;
 565
 566                                 f = io_u->file;
 567                                 if (io_u->offset == f->real_file_size)
 568                                         goto sync_done;
 569
 570                                 requeue_io_u(td, &io_u);
 571                         } else {
 572 sync_done:
 573                                 ret = io_u_sync_complete(td, io_u, bytes_done);
 574                                 if (ret < 0)
 575                                         break;
 576                         }
 577                         continue;
 578                 case FIO_Q_QUEUED:
 579                         break;
 580                 case FIO_Q_BUSY:
 581                         requeue_io_u(td, &io_u);
 582                         ret2 = td_io_commit(td);
 583                         if (ret2 < 0)
 584                                 ret = ret2;
 585                         break;
 586                 default:
 587                         assert(ret < 0);
 588                         td_verror(td, -ret, "td_io_queue");
 589                         break;
 590                 }
 591
 592                 if (break_on_this_error(td, ddir, &ret))
 593                         break;
 594
 595                 /*
 596                  * if we can queue more, do so. but check if there are
 597                  * completed io_u's first. Note that we can get BUSY even
 598                  * without IO queued, if the system is resource starved.
 599                  */
 600 reap:
 601                 full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
 602                 if (full || !td->o.iodepth_batch_complete) {
 603                         min_events = min(td->o.iodepth_batch_complete,
 604                                          td->cur_depth);
 605                         /*
 606                          * if the queue is full, we MUST reap at least 1 event
 607                          */
 608                         if (full && !min_events)
 609                                 min_events = 1;
 610
 611                         do {
 612                                 /*
 613                                  * Reap required number of io units, if any,
 614                                  * and do the verification on them through
 615                                  * the callback handler
 616                                  */
 617                                 if (io_u_queued_complete(td, min_events, bytes_done) < 0) {
 618                                         ret = -1;
 619                                         break;
 620                                 }
 621                         } while (full && (td->cur_depth > td->o.iodepth_low));
 622                 }
 623                 if (ret < 0)
 624                         break;
 625         }
 626
 627         check_update_rusage(td);
 628
 629         if (!td->error) {
 630                 min_events = td->cur_depth;
 631
 632                 if (min_events)
 633                         ret = io_u_queued_complete(td, min_events, NULL);
 634         } else
 635                 cleanup_pending_aio(td);
 636
 637         td_set_runstate(td, TD_RUNNING);
 638
 639         dprint(FD_VERIFY, "exiting loop\n");
 640 }
 641
 642 static unsigned int exceeds_number_ios(struct thread_data *td)
 643 {
 644         unsigned long long number_ios;
 645
 646         if (!td->o.number_ios)
 647                 return 0;
 648
 649         number_ios = ddir_rw_sum(td->this_io_blocks);
 650         number_ios += td->io_u_queued + td->io_u_in_flight;
 651
 652         return number_ios >= td->o.number_ios;
 653 }
 654
 655 static int io_bytes_exceeded(struct thread_data *td)
 656 {
 657         unsigned long long bytes, limit;
 658
 659         if (td_rw(td))
 660                 bytes = td->this_io_bytes[DDIR_READ] + td->this_io_bytes[DDIR_WRITE];
 661         else if (td_write(td))
 662                 bytes = td->this_io_bytes[DDIR_WRITE];
 663         else if (td_read(td))
 664                 bytes = td->this_io_bytes[DDIR_READ];
 665         else
 666                 bytes = td->this_io_bytes[DDIR_TRIM];
 667
 668         if (td->o.io_limit)
 669                 limit = td->o.io_limit;
 670         else
 671                 limit = td->o.size;
 672
 673         return bytes >= limit || exceeds_number_ios(td);
 674 }
 675
 676 /*
 677  * Main IO worker function. It retrieves io_u's to process and queues
 678  * and reaps them, checking for rate and errors along the way.
 679  *
 680  * Returns number of bytes written and trimmed.
 681  */
 682 static uint64_t do_io(struct thread_data *td)
 683 {
 684         uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 };
 685         unsigned int i;
 686         int ret = 0;
 687         uint64_t total_bytes, bytes_issued = 0;
 688
 689         if (in_ramp_time(td))
 690                 td_set_runstate(td, TD_RAMP);
 691         else
 692                 td_set_runstate(td, TD_RUNNING);
 693
 694         lat_target_init(td);
 695
 696         /*
 697          * If verify_backlog is enabled, we'll run the verify in this
 698          * handler as well. For that case, we may need up to twice the
 699          * amount of bytes.
 700          */
 701         total_bytes = td->o.size;
 702         if (td->o.verify != VERIFY_NONE &&
 703            (td_write(td) && td->o.verify_backlog))
 704                 total_bytes += td->o.size;
 705
 706         while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
 707                 (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) ||
 708                 td->o.time_based) {
 709                 struct timeval comp_time;
 710                 int min_evts = 0;
 711                 struct io_u *io_u;
 712                 int ret2, full;
 713                 enum fio_ddir ddir;
 714
 715                 check_update_rusage(td);
 716
 717                 if (td->terminate || td->done)
 718                         break;
 719
 720                 update_tv_cache(td);
 721
 722                 if (runtime_exceeded(td, &td->tv_cache)) {
 723                         __update_tv_cache(td);
 724                         if (runtime_exceeded(td, &td->tv_cache)) {
 725                                 fio_mark_td_terminate(td);
 726                                 break;
 727                         }
 728                 }
 729
 730                 if (flow_threshold_exceeded(td))
 731                         continue;
 732
 733                 if (bytes_issued >= total_bytes)
 734                         break;
 735
 736                 io_u = get_io_u(td);
 737                 if (IS_ERR_OR_NULL(io_u)) {
 738                         int err = PTR_ERR(io_u);
 739
 740                         io_u = NULL;
 741                         if (err == -EBUSY) {
 742                                 ret = FIO_Q_BUSY;
 743                                 goto reap;
 744                         }
 745                         if (td->o.latency_target)
 746                                 goto reap;
 747                         break;
 748                 }
 749
 750                 ddir = io_u->ddir;
 751
 752                 /*
 753                  * Add verification end_io handler if:
 754                  *      - Asked to verify (!td_rw(td))
 755                  *      - Or the io_u is from our verify list (mixed write/ver)
 756                  */
 757                 if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ &&
 758                     ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) {
 759
 760                         if (!td->o.verify_pattern_bytes) {
 761                                 io_u->rand_seed = __rand(&td->verify_state);
 762                                 if (sizeof(int) != sizeof(long *))
 763                                         io_u->rand_seed *= __rand(&td->verify_state);
 764                         }
 765
 766                         if (verify_state_should_stop(td, io_u)) {
 767                                 put_io_u(td, io_u);
 768                                 break;
 769                         }
 770
 771                         if (td->o.verify_async)
 772                                 io_u->end_io = verify_io_u_async;
 773                         else
 774                                 io_u->end_io = verify_io_u;
 775                         td_set_runstate(td, TD_VERIFYING);
 776                 } else if (in_ramp_time(td))
 777                         td_set_runstate(td, TD_RAMP);
 778                 else
 779                         td_set_runstate(td, TD_RUNNING);
 780
 781                 /*
 782                  * Always log IO before it's issued, so we know the specific
 783                  * order of it. The logged unit will track when the IO has
 784                  * completed.
 785                  */
 786                 if (td_write(td) && io_u->ddir == DDIR_WRITE &&
 787                     td->o.do_verify &&
 788                     td->o.verify != VERIFY_NONE &&
 789                     !td->o.experimental_verify)
 790                         log_io_piece(td, io_u);
 791
 792                 ret = td_io_queue(td, io_u);
 793                 switch (ret) {
 794                 case FIO_Q_COMPLETED:
 795                         if (io_u->error) {
 796                                 ret = -io_u->error;
 797                                 unlog_io_piece(td, io_u);
 798                                 clear_io_u(td, io_u);
 799                         } else if (io_u->resid) {
 800                                 int bytes = io_u->xfer_buflen - io_u->resid;
 801                                 struct fio_file *f = io_u->file;
 802
 803                                 bytes_issued += bytes;
 804
 805                                 trim_io_piece(td, io_u);
 806
 807                                 /*
 808                                  * zero read, fail
 809                                  */
 810                                 if (!bytes) {
 811                                         unlog_io_piece(td, io_u);
 812                                         td_verror(td, EIO, "full resid");
 813                                         put_io_u(td, io_u);
 814                                         break;
 815                                 }
 816
 817                                 io_u->xfer_buflen = io_u->resid;
 818                                 io_u->xfer_buf += bytes;
 819                                 io_u->offset += bytes;
 820
 821                                 if (ddir_rw(io_u->ddir))
 822                                         td->ts.short_io_u[io_u->ddir]++;
 823
 824                                 if (io_u->offset == f->real_file_size)
 825                                         goto sync_done;
 826
 827                                 requeue_io_u(td, &io_u);
 828                         } else {
 829 sync_done:
 830                                 if (__should_check_rate(td, DDIR_READ) ||
 831                                     __should_check_rate(td, DDIR_WRITE) ||
 832                                     __should_check_rate(td, DDIR_TRIM))
 833                                         fio_gettime(&comp_time, NULL);
 834
 835                                 ret = io_u_sync_complete(td, io_u, bytes_done);
 836                                 if (ret < 0)
 837                                         break;
 838                                 bytes_issued += io_u->xfer_buflen;
 839                         }
 840                         break;
 841                 case FIO_Q_QUEUED:
 842                         /*
 843                          * if the engine doesn't have a commit hook,
 844                          * the io_u is really queued. if it does have such
 845                          * a hook, it has to call io_u_queued() itself.
 846                          */
 847                         if (td->io_ops->commit == NULL)
 848                                 io_u_queued(td, io_u);
 849                         bytes_issued += io_u->xfer_buflen;
 850                         break;
 851                 case FIO_Q_BUSY:
 852                         unlog_io_piece(td, io_u);
 853                         requeue_io_u(td, &io_u);
 854                         ret2 = td_io_commit(td);
 855                         if (ret2 < 0)
 856                                 ret = ret2;
 857                         break;
 858                 default:
 859                         assert(ret < 0);
 860                         put_io_u(td, io_u);
 861                         break;
 862                 }
 863
 864                 if (break_on_this_error(td, ddir, &ret))
 865                         break;
 866
 867                 /*
 868                  * See if we need to complete some commands. Note that we
 869                  * can get BUSY even without IO queued, if the system is
 870                  * resource starved.
 871                  */
 872 reap:
 873                 full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
 874                 if (full || !td->o.iodepth_batch_complete) {
 875                         min_evts = min(td->o.iodepth_batch_complete,
 876                                         td->cur_depth);
 877                         /*
 878                          * if the queue is full, we MUST reap at least 1 event
 879                          */
 880                         if (full && !min_evts)
 881                                 min_evts = 1;
 882
 883                         if (__should_check_rate(td, DDIR_READ) ||
 884                             __should_check_rate(td, DDIR_WRITE) ||
 885                             __should_check_rate(td, DDIR_TRIM))
 886                                 fio_gettime(&comp_time, NULL);
 887
 888                         do {
 889                                 ret = io_u_queued_complete(td, min_evts, bytes_done);
 890                                 if (ret < 0)
 891                                         break;
 892
 893                         } while (full && (td->cur_depth > td->o.iodepth_low));
 894                 }
 895
 896                 if (ret < 0)
 897                         break;
 898                 if (!ddir_rw_sum(bytes_done) && !(td->io_ops->flags & FIO_NOIO))
 899                         continue;
 900
 901                 if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) {
 902                         if (check_min_rate(td, &comp_time, bytes_done)) {
 903                                 if (exitall_on_terminate)
 904                                         fio_terminate_threads(td->groupid);
 905                                 td_verror(td, EIO, "check_min_rate");
 906                                 break;
 907                         }
 908                 }
 909                 if (!in_ramp_time(td) && td->o.latency_target)
 910                         lat_target_check(td);
 911
 912                 if (td->o.thinktime) {
 913                         unsigned long long b;
 914
 915                         b = ddir_rw_sum(td->io_blocks);
 916                         if (!(b % td->o.thinktime_blocks)) {
 917                                 int left;
 918
 919                                 io_u_quiesce(td);
 920
 921                                 if (td->o.thinktime_spin)
 922                                         usec_spin(td->o.thinktime_spin);
 923
 924                                 left = td->o.thinktime - td->o.thinktime_spin;
 925                                 if (left)
 926                                         usec_sleep(td, left);
 927                         }
 928                 }
 929         }
 930
 931         check_update_rusage(td);
 932
 933         if (td->trim_entries)
 934                 log_err("fio: %lu trim entries leaked?\n", td->trim_entries);
 935
 936         if (td->o.fill_device && td->error == ENOSPC) {
 937                 td->error = 0;
 938                 fio_mark_td_terminate(td);
 939         }
 940         if (!td->error) {
 941                 struct fio_file *f;
 942
 943                 i = td->cur_depth;
 944                 if (i) {
 945                         ret = io_u_queued_complete(td, i, bytes_done);
 946                         if (td->o.fill_device && td->error == ENOSPC)
 947                                 td->error = 0;
 948                 }
 949
 950                 if (should_fsync(td) && td->o.end_fsync) {
 951                         td_set_runstate(td, TD_FSYNCING);
 952
 953                         for_each_file(td, f, i) {
 954                                 if (!fio_file_fsync(td, f))
 955                                         continue;
 956
 957                                 log_err("fio: end_fsync failed for file %s\n",
 958                                                                 f->file_name);
 959                         }
 960                 }
 961         } else
 962                 cleanup_pending_aio(td);
 963
 964         /*
 965          * stop job if we failed doing any IO
 966          */
 967         if (!ddir_rw_sum(td->this_io_bytes))
 968                 td->done = 1;
 969
 970         return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM];
 971 }
 972
 973 static void cleanup_io_u(struct thread_data *td)
 974 {
 975         struct io_u *io_u;
 976
 977         while ((io_u = io_u_qpop(&td->io_u_freelist)) != NULL) {
 978
 979                 if (td->io_ops->io_u_free)
 980                         td->io_ops->io_u_free(td, io_u);
 981
 982                 fio_memfree(io_u, sizeof(*io_u));
 983         }
 984
 985         free_io_mem(td);
 986
 987         io_u_rexit(&td->io_u_requeues);
 988         io_u_qexit(&td->io_u_freelist);
 989         io_u_qexit(&td->io_u_all);
 990
 991         if (td->last_write_comp)
 992                 sfree(td->last_write_comp);
 993 }
 994
 995 static int init_io_u(struct thread_data *td)
 996 {
 997         struct io_u *io_u;
 998         unsigned int max_bs, min_write;
 999         int cl_align, i, max_units;
1000         int data_xfer = 1, err;
1001         char *p;
1002
1003         max_units = td->o.iodepth;
1004         max_bs = td_max_bs(td);
1005         min_write = td->o.min_bs[DDIR_WRITE];
1006         td->orig_buffer_size = (unsigned long long) max_bs
1007                                         * (unsigned long long) max_units;
1008
1009         if ((td->io_ops->flags & FIO_NOIO) || !(td_read(td) || td_write(td)))
1010                 data_xfer = 0;
1011
1012         err = 0;
1013         err += io_u_rinit(&td->io_u_requeues, td->o.iodepth);
1014         err += io_u_qinit(&td->io_u_freelist, td->o.iodepth);
1015         err += io_u_qinit(&td->io_u_all, td->o.iodepth);
1016
1017         if (err) {
1018                 log_err("fio: failed setting up IO queues\n");
1019                 return 1;
1020         }
1021
1022         /*
1023          * if we may later need to do address alignment, then add any
1024          * possible adjustment here so that we don't cause a buffer
1025          * overflow later. this adjustment may be too much if we get
1026          * lucky and the allocator gives us an aligned address.
1027          */
1028         if (td->o.odirect || td->o.mem_align || td->o.oatomic ||
1029             (td->io_ops->flags & FIO_RAWIO))
1030                 td->orig_buffer_size += page_mask + td->o.mem_align;
1031
1032         if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) {
1033                 unsigned long bs;
1034
1035                 bs = td->orig_buffer_size + td->o.hugepage_size - 1;
1036                 td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1);
1037         }
1038
1039         if (td->orig_buffer_size != (size_t) td->orig_buffer_size) {
1040                 log_err("fio: IO memory too large. Reduce max_bs or iodepth\n");
1041                 return 1;
1042         }
1043
1044         if (data_xfer && allocate_io_mem(td))
1045                 return 1;
1046
1047         if (td->o.odirect || td->o.mem_align || td->o.oatomic ||
1048             (td->io_ops->flags & FIO_RAWIO))
1049                 p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align;
1050         else
1051                 p = td->orig_buffer;
1052
1053         cl_align = os_cache_line_size();
1054
1055         for (i = 0; i < max_units; i++) {
1056                 void *ptr;
1057
1058                 if (td->terminate)
1059                         return 1;
1060
1061                 ptr = fio_memalign(cl_align, sizeof(*io_u));
1062                 if (!ptr) {
1063                         log_err("fio: unable to allocate aligned memory\n");
1064                         break;
1065                 }
1066
1067                 io_u = ptr;
1068                 memset(io_u, 0, sizeof(*io_u));
1069                 INIT_FLIST_HEAD(&io_u->verify_list);
1070                 dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i);
1071
1072                 if (data_xfer) {
1073                         io_u->buf = p;
1074                         dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf);
1075
1076                         if (td_write(td))
1077                                 io_u_fill_buffer(td, io_u, min_write, max_bs);
1078                         if (td_write(td) && td->o.verify_pattern_bytes) {
1079                                 /*
1080                                  * Fill the buffer with the pattern if we are
1081                                  * going to be doing writes.
1082                                  */
1083                                 fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0);
1084                         }
1085                 }
1086
1087                 io_u->index = i;
1088                 io_u->flags = IO_U_F_FREE;
1089                 io_u_qpush(&td->io_u_freelist, io_u);
1090
1091                 /*
1092                  * io_u never leaves this stack, used for iteration of all
1093                  * io_u buffers.
1094                  */
1095                 io_u_qpush(&td->io_u_all, io_u);
1096
1097                 if (td->io_ops->io_u_init) {
1098                         int ret = td->io_ops->io_u_init(td, io_u);
1099
1100                         if (ret) {
1101                                 log_err("fio: failed to init engine data: %d\n", ret);
1102                                 return 1;
1103                         }
1104                 }
1105
1106                 p += max_bs;
1107         }
1108
1109         if (td->o.verify != VERIFY_NONE) {
1110                 td->last_write_comp = scalloc(max_units, sizeof(uint64_t));
1111                 if (!td->last_write_comp) {
1112                         log_err("fio: failed to alloc write comp data\n");
1113                         return 1;
1114                 }
1115         }
1116
1117         return 0;
1118 }
1119
1120 static int switch_ioscheduler(struct thread_data *td)
1121 {
1122         char tmp[256], tmp2[128];
1123         FILE *f;
1124         int ret;
1125
1126         if (td->io_ops->flags & FIO_DISKLESSIO)
1127                 return 0;
1128
1129         sprintf(tmp, "%s/queue/scheduler", td->sysfs_root);
1130
1131         f = fopen(tmp, "r+");
1132         if (!f) {
1133                 if (errno == ENOENT) {
1134                         log_err("fio: os or kernel doesn't support IO scheduler"
1135                                 " switching\n");
1136                         return 0;
1137                 }
1138                 td_verror(td, errno, "fopen iosched");
1139                 return 1;
1140         }
1141
1142         /*
1143          * Set io scheduler.
1144          */
1145         ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f);
1146         if (ferror(f) || ret != 1) {
1147                 td_verror(td, errno, "fwrite");
1148                 fclose(f);
1149                 return 1;
1150         }
1151
1152         rewind(f);
1153
1154         /*
1155          * Read back and check that the selected scheduler is now the default.
1156          */
1157         ret = fread(tmp, sizeof(tmp), 1, f);
1158         if (ferror(f) || ret < 0) {
1159                 td_verror(td, errno, "fread");
1160                 fclose(f);
1161                 return 1;
1162         }
1163         tmp[sizeof(tmp) - 1] = '\0';
1164
1165
1166         sprintf(tmp2, "[%s]", td->o.ioscheduler);
1167         if (!strstr(tmp, tmp2)) {
1168                 log_err("fio: io scheduler %s not found\n", td->o.ioscheduler);
1169                 td_verror(td, EINVAL, "iosched_switch");
1170                 fclose(f);
1171                 return 1;
1172         }
1173
1174         fclose(f);
1175         return 0;
1176 }
1177
1178 static int keep_running(struct thread_data *td)
1179 {
1180         unsigned long long limit;
1181
1182         if (td->done)
1183                 return 0;
1184         if (td->o.time_based)
1185                 return 1;
1186         if (td->o.loops) {
1187                 td->o.loops--;
1188                 return 1;
1189         }
1190         if (exceeds_number_ios(td))
1191                 return 0;
1192
1193         if (td->o.io_limit)
1194                 limit = td->o.io_limit;
1195         else
1196                 limit = td->o.size;
1197
1198         if (limit != -1ULL && ddir_rw_sum(td->io_bytes) < limit) {
1199                 uint64_t diff;
1200
1201                 /*
1202                  * If the difference is less than the minimum IO size, we
1203                  * are done.
1204                  */
1205                 diff = limit - ddir_rw_sum(td->io_bytes);
1206                 if (diff < td_max_bs(td))
1207                         return 0;
1208
1209                 if (fio_files_done(td))
1210                         return 0;
1211
1212                 return 1;
1213         }
1214
1215         return 0;
1216 }
1217
1218 static int exec_string(struct thread_options *o, const char *string, const char *mode)
1219 {
1220         int ret, newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1;
1221         char *str;
1222
1223         str = malloc(newlen);
1224         sprintf(str, "%s &> %s.%s.txt", string, o->name, mode);
1225
1226         log_info("%s : Saving output of %s in %s.%s.txt\n",o->name, mode, o->name, mode);
1227         ret = system(str);
1228         if (ret == -1)
1229                 log_err("fio: exec of cmd <%s> failed\n", str);
1230
1231         free(str);
1232         return ret;
1233 }
1234
1235 /*
1236  * Dry run to compute correct state of numberio for verification.
1237  */
1238 static uint64_t do_dry_run(struct thread_data *td)
1239 {
1240         uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 };
1241
1242         td_set_runstate(td, TD_RUNNING);
1243
1244         while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
1245                 (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td)) {
1246                 struct io_u *io_u;
1247                 int ret;
1248
1249                 if (td->terminate || td->done)
1250                         break;
1251
1252                 io_u = get_io_u(td);
1253                 if (!io_u)
1254                         break;
1255
1256                 io_u->flags |= IO_U_F_FLIGHT;
1257                 io_u->error = 0;
1258                 io_u->resid = 0;
1259                 if (ddir_rw(acct_ddir(io_u)))
1260                         td->io_issues[acct_ddir(io_u)]++;
1261                 if (ddir_rw(io_u->ddir)) {
1262                         io_u_mark_depth(td, 1);
1263                         td->ts.total_io_u[io_u->ddir]++;
1264                 }
1265
1266                 if (td_write(td) && io_u->ddir == DDIR_WRITE &&
1267                     td->o.do_verify &&
1268                     td->o.verify != VERIFY_NONE &&
1269                     !td->o.experimental_verify)
1270                         log_io_piece(td, io_u);
1271
1272                 ret = io_u_sync_complete(td, io_u, bytes_done);
1273                 (void) ret;
1274         }
1275
1276         return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM];
1277 }
1278
1279 /*
1280  * Entry point for the thread based jobs. The process based jobs end up
1281  * here as well, after a little setup.
1282  */
1283 static void *thread_main(void *data)
1284 {
1285         unsigned long long elapsed;
1286         struct thread_data *td = data;
1287         struct thread_options *o = &td->o;
1288         pthread_condattr_t attr;
1289         int clear_state;
1290         int ret;
1291
1292         if (!o->use_thread) {
1293                 setsid();
1294                 td->pid = getpid();
1295         } else
1296                 td->pid = gettid();
1297
1298         fio_local_clock_init(o->use_thread);
1299
1300         dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid);
1301
1302         if (is_backend)
1303                 fio_server_send_start(td);
1304
1305         INIT_FLIST_HEAD(&td->io_log_list);
1306         INIT_FLIST_HEAD(&td->io_hist_list);
1307         INIT_FLIST_HEAD(&td->verify_list);
1308         INIT_FLIST_HEAD(&td->trim_list);
1309         INIT_FLIST_HEAD(&td->next_rand_list);
1310         pthread_mutex_init(&td->io_u_lock, NULL);
1311         td->io_hist_tree = RB_ROOT;
1312
1313         pthread_condattr_init(&attr);
1314         pthread_cond_init(&td->verify_cond, &attr);
1315         pthread_cond_init(&td->free_cond, &attr);
1316
1317         td_set_runstate(td, TD_INITIALIZED);
1318         dprint(FD_MUTEX, "up startup_mutex\n");
1319         fio_mutex_up(startup_mutex);
1320         dprint(FD_MUTEX, "wait on td->mutex\n");
1321         fio_mutex_down(td->mutex);
1322         dprint(FD_MUTEX, "done waiting on td->mutex\n");
1323
1324         /*
1325          * A new gid requires privilege, so we need to do this before setting
1326          * the uid.
1327          */
1328         if (o->gid != -1U && setgid(o->gid)) {
1329                 td_verror(td, errno, "setgid");
1330                 goto err;
1331         }
1332         if (o->uid != -1U && setuid(o->uid)) {
1333                 td_verror(td, errno, "setuid");
1334                 goto err;
1335         }
1336
1337         /*
1338          * If we have a gettimeofday() thread, make sure we exclude that
1339          * thread from this job
1340          */
1341         if (o->gtod_cpu)
1342                 fio_cpu_clear(&o->cpumask, o->gtod_cpu);
1343
1344         /*
1345          * Set affinity first, in case it has an impact on the memory
1346          * allocations.
1347          */
1348         if (o->cpumask_set) {
1349                 if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) {
1350                         ret = fio_cpus_split(&o->cpumask, td->thread_number - 1);
1351                         if (!ret) {
1352                                 log_err("fio: no CPUs set\n");
1353                                 log_err("fio: Try increasing number of available CPUs\n");
1354                                 td_verror(td, EINVAL, "cpus_split");
1355                                 goto err;
1356                         }
1357                 }
1358                 ret = fio_setaffinity(td->pid, o->cpumask);
1359                 if (ret == -1) {
1360                         td_verror(td, errno, "cpu_set_affinity");
1361                         goto err;
1362                 }
1363         }
1364
1365 #ifdef CONFIG_LIBNUMA
1366         /* numa node setup */
1367         if (o->numa_cpumask_set || o->numa_memmask_set) {
1368                 struct bitmask *mask;
1369
1370                 if (numa_available() < 0) {
1371                         td_verror(td, errno, "Does not support NUMA API\n");
1372                         goto err;
1373                 }
1374
1375                 if (o->numa_cpumask_set) {
1376                         mask = numa_parse_nodestring(o->numa_cpunodes);
1377                         ret = numa_run_on_node_mask(mask);
1378                         numa_free_nodemask(mask);
1379                         if (ret == -1) {
1380                                 td_verror(td, errno, \
1381                                         "numa_run_on_node_mask failed\n");
1382                                 goto err;
1383                         }
1384                 }
1385
1386                 if (o->numa_memmask_set) {
1387
1388                         mask = NULL;
1389                         if (o->numa_memnodes)
1390                                 mask = numa_parse_nodestring(o->numa_memnodes);
1391
1392                         switch (o->numa_mem_mode) {
1393                         case MPOL_INTERLEAVE:
1394                                 numa_set_interleave_mask(mask);
1395                                 break;
1396                         case MPOL_BIND:
1397                                 numa_set_membind(mask);
1398                                 break;
1399                         case MPOL_LOCAL:
1400                                 numa_set_localalloc();
1401                                 break;
1402                         case MPOL_PREFERRED:
1403                                 numa_set_preferred(o->numa_mem_prefer_node);
1404                                 break;
1405                         case MPOL_DEFAULT:
1406                         default:
1407                                 break;
1408                         }
1409
1410                         if (mask)
1411                                 numa_free_nodemask(mask);
1412
1413                 }
1414         }
1415 #endif
1416
1417         if (fio_pin_memory(td))
1418                 goto err;
1419
1420         /*
1421          * May alter parameters that init_io_u() will use, so we need to
1422          * do this first.
1423          */
1424         if (init_iolog(td))
1425                 goto err;
1426
1427         if (init_io_u(td))
1428                 goto err;
1429
1430         if (o->verify_async && verify_async_init(td))
1431                 goto err;
1432
1433         if (o->ioprio) {
1434                 ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio);
1435                 if (ret == -1) {
1436                         td_verror(td, errno, "ioprio_set");
1437                         goto err;
1438                 }
1439         }
1440
1441         if (o->cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt))
1442                 goto err;
1443
1444         errno = 0;
1445         if (nice(o->nice) == -1 && errno != 0) {
1446                 td_verror(td, errno, "nice");
1447                 goto err;
1448         }
1449
1450         if (o->ioscheduler && switch_ioscheduler(td))
1451                 goto err;
1452
1453         if (!o->create_serialize && setup_files(td))
1454                 goto err;
1455
1456         if (td_io_init(td))
1457                 goto err;
1458
1459         if (init_random_map(td))
1460                 goto err;
1461
1462         if (o->exec_prerun && exec_string(o, o->exec_prerun, (const char *)"prerun"))
1463                 goto err;
1464
1465         if (o->pre_read) {
1466                 if (pre_read_files(td) < 0)
1467                         goto err;
1468         }
1469
1470         if (td->flags & TD_F_COMPRESS_LOG)
1471                 tp_init(&td->tp_data);
1472
1473         fio_verify_init(td);
1474
1475         fio_gettime(&td->epoch, NULL);
1476         fio_getrusage(&td->ru_start);
1477         clear_state = 0;
1478         while (keep_running(td)) {
1479                 uint64_t verify_bytes;
1480
1481                 fio_gettime(&td->start, NULL);
1482                 memcpy(&td->bw_sample_time, &td->start, sizeof(td->start));
1483                 memcpy(&td->iops_sample_time, &td->start, sizeof(td->start));
1484                 memcpy(&td->tv_cache, &td->start, sizeof(td->start));
1485
1486                 if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] ||
1487                                 o->ratemin[DDIR_TRIM]) {
1488                         memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time,
1489                                                 sizeof(td->bw_sample_time));
1490                         memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time,
1491                                                 sizeof(td->bw_sample_time));
1492                         memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time,
1493                                                 sizeof(td->bw_sample_time));
1494                 }
1495
1496                 if (clear_state)
1497                         clear_io_state(td);
1498
1499                 prune_io_piece_log(td);
1500
1501                 if (td->o.verify_only && (td_write(td) || td_rw(td)))
1502                         verify_bytes = do_dry_run(td);
1503                 else
1504                         verify_bytes = do_io(td);
1505
1506                 clear_state = 1;
1507
1508                 fio_mutex_down(stat_mutex);
1509                 if (td_read(td) && td->io_bytes[DDIR_READ]) {
1510                         elapsed = mtime_since_now(&td->start);
1511                         td->ts.runtime[DDIR_READ] += elapsed;
1512                 }
1513                 if (td_write(td) && td->io_bytes[DDIR_WRITE]) {
1514                         elapsed = mtime_since_now(&td->start);
1515                         td->ts.runtime[DDIR_WRITE] += elapsed;
1516                 }
1517                 if (td_trim(td) && td->io_bytes[DDIR_TRIM]) {
1518                         elapsed = mtime_since_now(&td->start);
1519                         td->ts.runtime[DDIR_TRIM] += elapsed;
1520                 }
1521                 fio_gettime(&td->start, NULL);
1522                 fio_mutex_up(stat_mutex);
1523
1524                 if (td->error || td->terminate)
1525                         break;
1526
1527                 if (!o->do_verify ||
1528                     o->verify == VERIFY_NONE ||
1529                     (td->io_ops->flags & FIO_UNIDIR))
1530                         continue;
1531
1532                 clear_io_state(td);
1533
1534                 fio_gettime(&td->start, NULL);
1535
1536                 do_verify(td, verify_bytes);
1537
1538                 fio_mutex_down(stat_mutex);
1539                 td->ts.runtime[DDIR_READ] += mtime_since_now(&td->start);
1540                 fio_gettime(&td->start, NULL);
1541                 fio_mutex_up(stat_mutex);
1542
1543                 if (td->error || td->terminate)
1544                         break;
1545         }
1546
1547         update_rusage_stat(td);
1548         td->ts.total_run_time = mtime_since_now(&td->epoch);
1549         td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ];
1550         td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE];
1551         td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM];
1552
1553         if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) &&
1554             (td->o.verify != VERIFY_NONE && td_write(td))) {
1555                 struct all_io_list *state;
1556                 size_t sz;
1557
1558                 state = get_all_io_list(td->thread_number, &sz);
1559                 if (state) {
1560                         __verify_save_state(state, "local");
1561                         free(state);
1562                 }
1563         }
1564
1565         fio_unpin_memory(td);
1566
1567         fio_writeout_logs(td);
1568
1569         if (td->flags & TD_F_COMPRESS_LOG)
1570                 tp_exit(&td->tp_data);
1571
1572         if (o->exec_postrun)
1573                 exec_string(o, o->exec_postrun, (const char *)"postrun");
1574
1575         if (exitall_on_terminate)
1576                 fio_terminate_threads(td->groupid);
1577
1578 err:
1579         if (td->error)
1580                 log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error,
1581                                                         td->verror);
1582
1583         if (o->verify_async)
1584                 verify_async_exit(td);
1585
1586         close_and_free_files(td);
1587         cleanup_io_u(td);
1588         close_ioengine(td);
1589         cgroup_shutdown(td, &cgroup_mnt);
1590         verify_free_state(td);
1591
1592         if (o->cpumask_set) {
1593                 ret = fio_cpuset_exit(&o->cpumask);
1594                 if (ret)
1595                         td_verror(td, ret, "fio_cpuset_exit");
1596         }
1597
1598         /*
1599          * do this very late, it will log file closing as well
1600          */
1601         if (o->write_iolog_file)
1602                 write_iolog_close(td);
1603
1604         fio_mutex_remove(td->mutex);
1605         td->mutex = NULL;
1606
1607         td_set_runstate(td, TD_EXITED);
1608
1609         /*
1610          * Do this last after setting our runstate to exited, so we
1611          * know that the stat thread is signaled.
1612          */
1613         check_update_rusage(td);
1614
1615         return (void *) (uintptr_t) td->error;
1616 }
1617
1618
1619 /*
1620  * We cannot pass the td data into a forked process, so attach the td and
1621  * pass it to the thread worker.
1622  */
1623 static int fork_main(int shmid, int offset)
1624 {
1625         struct thread_data *td;
1626         void *data, *ret;
1627
1628 #if !defined(__hpux) && !defined(CONFIG_NO_SHM)
1629         data = shmat(shmid, NULL, 0);
1630         if (data == (void *) -1) {
1631                 int __err = errno;
1632
1633                 perror("shmat");
1634                 return __err;
1635         }
1636 #else
1637         /*
1638          * HP-UX inherits shm mappings?
1639          */
1640         data = threads;
1641 #endif
1642
1643         td = data + offset * sizeof(struct thread_data);
1644         ret = thread_main(td);
1645         shmdt(data);
1646         return (int) (uintptr_t) ret;
1647 }
1648
1649 static void dump_td_info(struct thread_data *td)
1650 {
1651         log_err("fio: job '%s' hasn't exited in %lu seconds, it appears to "
1652                 "be stuck. Doing forceful exit of this job.\n", td->o.name,
1653                         (unsigned long) time_since_now(&td->terminate_time));
1654 }
1655
1656 /*
1657  * Run over the job map and reap the threads that have exited, if any.
1658  */
1659 static void reap_threads(unsigned int *nr_running, unsigned int *t_rate,
1660                          unsigned int *m_rate)
1661 {
1662         struct thread_data *td;
1663         unsigned int cputhreads, realthreads, pending;
1664         int i, status, ret;
1665
1666         /*
1667          * reap exited threads (TD_EXITED -> TD_REAPED)
1668          */
1669         realthreads = pending = cputhreads = 0;
1670         for_each_td(td, i) {
1671                 int flags = 0;
1672
1673                 /*
1674                  * ->io_ops is NULL for a thread that has closed its
1675                  * io engine
1676                  */
1677                 if (td->io_ops && !strcmp(td->io_ops->name, "cpuio"))
1678                         cputhreads++;
1679                 else
1680                         realthreads++;
1681
1682                 if (!td->pid) {
1683                         pending++;
1684                         continue;
1685                 }
1686                 if (td->runstate == TD_REAPED)
1687                         continue;
1688                 if (td->o.use_thread) {
1689                         if (td->runstate == TD_EXITED) {
1690                                 td_set_runstate(td, TD_REAPED);
1691                                 goto reaped;
1692                         }
1693                         continue;
1694                 }
1695
1696                 flags = WNOHANG;
1697                 if (td->runstate == TD_EXITED)
1698                         flags = 0;
1699
1700                 /*
1701                  * check if someone quit or got killed in an unusual way
1702                  */
1703                 ret = waitpid(td->pid, &status, flags);
1704                 if (ret < 0) {
1705                         if (errno == ECHILD) {
1706                                 log_err("fio: pid=%d disappeared %d\n",
1707                                                 (int) td->pid, td->runstate);
1708                                 td->sig = ECHILD;
1709                                 td_set_runstate(td, TD_REAPED);
1710                                 goto reaped;
1711                         }
1712                         perror("waitpid");
1713                 } else if (ret == td->pid) {
1714                         if (WIFSIGNALED(status)) {
1715                                 int sig = WTERMSIG(status);
1716
1717                                 if (sig != SIGTERM && sig != SIGUSR2)
1718                                         log_err("fio: pid=%d, got signal=%d\n",
1719                                                         (int) td->pid, sig);
1720                                 td->sig = sig;
1721                                 td_set_runstate(td, TD_REAPED);
1722                                 goto reaped;
1723                         }
1724                         if (WIFEXITED(status)) {
1725                                 if (WEXITSTATUS(status) && !td->error)
1726                                         td->error = WEXITSTATUS(status);
1727
1728                                 td_set_runstate(td, TD_REAPED);
1729                                 goto reaped;
1730                         }
1731                 }
1732
1733                 /*
1734                  * If the job is stuck, do a forceful timeout of it and
1735                  * move on.
1736                  */
1737                 if (td->terminate &&
1738                     time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) {
1739                         dump_td_info(td);
1740                         td_set_runstate(td, TD_REAPED);
1741                         goto reaped;
1742                 }
1743
1744                 /*
1745                  * thread is not dead, continue
1746                  */
1747                 pending++;
1748                 continue;
1749 reaped:
1750                 (*nr_running)--;
1751                 (*m_rate) -= ddir_rw_sum(td->o.ratemin);
1752                 (*t_rate) -= ddir_rw_sum(td->o.rate);
1753                 if (!td->pid)
1754                         pending--;
1755
1756                 if (td->error)
1757                         exit_value++;
1758
1759                 done_secs += mtime_since_now(&td->epoch) / 1000;
1760                 profile_td_exit(td);
1761         }
1762
1763         if (*nr_running == cputhreads && !pending && realthreads)
1764                 fio_terminate_threads(TERMINATE_ALL);
1765 }
1766
1767 static int __check_trigger_file(void)
1768 {
1769         struct stat sb;
1770
1771         if (!trigger_file)
1772                 return 0;
1773
1774         if (stat(trigger_file, &sb))
1775                 return 0;
1776
1777         if (unlink(trigger_file) < 0)
1778                 log_err("fio: failed to unlink %s: %s\n", trigger_file,
1779                                                         strerror(errno));
1780
1781         return 1;
1782 }
1783
1784 static int trigger_timedout(void)
1785 {
1786         if (trigger_timeout)
1787                 return time_since_genesis() >= trigger_timeout;
1788
1789         return 0;
1790 }
1791
1792 void exec_trigger(const char *cmd)
1793 {
1794         int ret;
1795
1796         if (!cmd)
1797                 return;
1798
1799         ret = system(cmd);
1800         if (ret == -1)
1801                 log_err("fio: failed executing %s trigger\n", cmd);
1802 }
1803
1804 void check_trigger_file(void)
1805 {
1806         if (__check_trigger_file() || trigger_timedout()) {
1807                 if (nr_clients) {
1808                         if (trigger_remote_cmd)
1809                                 fio_clients_send_trigger(trigger_remote_cmd);
1810                 } else {
1811                         verify_save_state();
1812                         fio_terminate_threads(TERMINATE_ALL);
1813                         exec_trigger(trigger_cmd);
1814                 }
1815         }
1816 }
1817
1818 static int fio_verify_load_state(struct thread_data *td)
1819 {
1820         int ret;
1821
1822         if (!td->o.verify_state)
1823                 return 0;
1824
1825         if (is_backend) {
1826                 void *data;
1827
1828                 ret = fio_server_get_verify_state(td->o.name,
1829                                         td->thread_number - 1, &data);
1830                 if (!ret)
1831                         verify_convert_assign_state(td, data);
1832         } else
1833                 ret = verify_load_state(td, "local");
1834
1835         return ret;
1836 }
1837
1838 static void do_usleep(unsigned int usecs)
1839 {
1840         check_for_running_stats();
1841         check_trigger_file();
1842         usleep(usecs);
1843 }
1844
1845 /*
1846  * Main function for kicking off and reaping jobs, as needed.
1847  */
1848 static void run_threads(void)
1849 {
1850         struct thread_data *td;
1851         unsigned int i, todo, nr_running, m_rate, t_rate, nr_started;
1852         uint64_t spent;
1853
1854         if (fio_gtod_offload && fio_start_gtod_thread())
1855                 return;
1856
1857         fio_idle_prof_init();
1858
1859         set_sig_handlers();
1860
1861         nr_thread = nr_process = 0;
1862         for_each_td(td, i) {
1863                 if (td->o.use_thread)
1864                         nr_thread++;
1865                 else
1866                         nr_process++;
1867         }
1868
1869         if (output_format == FIO_OUTPUT_NORMAL) {
1870                 log_info("Starting ");
1871                 if (nr_thread)
1872                         log_info("%d thread%s", nr_thread,
1873                                                 nr_thread > 1 ? "s" : "");
1874                 if (nr_process) {
1875                         if (nr_thread)
1876                                 log_info(" and ");
1877                         log_info("%d process%s", nr_process,
1878                                                 nr_process > 1 ? "es" : "");
1879                 }
1880                 log_info("\n");
1881                 log_info_flush();
1882         }
1883
1884         todo = thread_number;
1885         nr_running = 0;
1886         nr_started = 0;
1887         m_rate = t_rate = 0;
1888
1889         for_each_td(td, i) {
1890                 print_status_init(td->thread_number - 1);
1891
1892                 if (!td->o.create_serialize)
1893                         continue;
1894
1895                 if (fio_verify_load_state(td))
1896                         goto reap;
1897
1898                 /*
1899                  * do file setup here so it happens sequentially,
1900                  * we don't want X number of threads getting their
1901                  * client data interspersed on disk
1902                  */
1903                 if (setup_files(td)) {
1904 reap:
1905                         exit_value++;
1906                         if (td->error)
1907                                 log_err("fio: pid=%d, err=%d/%s\n",
1908                                         (int) td->pid, td->error, td->verror);
1909                         td_set_runstate(td, TD_REAPED);
1910                         todo--;
1911                 } else {
1912                         struct fio_file *f;
1913                         unsigned int j;
1914
1915                         /*
1916                          * for sharing to work, each job must always open
1917                          * its own files. so close them, if we opened them
1918                          * for creation
1919                          */
1920                         for_each_file(td, f, j) {
1921                                 if (fio_file_open(f))
1922                                         td_io_close_file(td, f);
1923                         }
1924                 }
1925         }
1926
1927         /* start idle threads before io threads start to run */
1928         fio_idle_prof_start();
1929
1930         set_genesis_time();
1931
1932         while (todo) {
1933                 struct thread_data *map[REAL_MAX_JOBS];
1934                 struct timeval this_start;
1935                 int this_jobs = 0, left;
1936
1937                 /*
1938                  * create threads (TD_NOT_CREATED -> TD_CREATED)
1939                  */
1940                 for_each_td(td, i) {
1941                         if (td->runstate != TD_NOT_CREATED)
1942                                 continue;
1943
1944                         /*
1945                          * never got a chance to start, killed by other
1946                          * thread for some reason
1947                          */
1948                         if (td->terminate) {
1949                                 todo--;
1950                                 continue;
1951                         }
1952
1953                         if (td->o.start_delay) {
1954                                 spent = utime_since_genesis();
1955
1956                                 if (td->o.start_delay > spent)
1957                                         continue;
1958                         }
1959
1960                         if (td->o.stonewall && (nr_started || nr_running)) {
1961                                 dprint(FD_PROCESS, "%s: stonewall wait\n",
1962                                                         td->o.name);
1963                                 break;
1964                         }
1965
1966                         init_disk_util(td);
1967
1968                         td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED);
1969                         td->update_rusage = 0;
1970
1971                         /*
1972                          * Set state to created. Thread will transition
1973                          * to TD_INITIALIZED when it's done setting up.
1974                          */
1975                         td_set_runstate(td, TD_CREATED);
1976                         map[this_jobs++] = td;
1977                         nr_started++;
1978
1979                         if (td->o.use_thread) {
1980                                 int ret;
1981
1982                                 dprint(FD_PROCESS, "will pthread_create\n");
1983                                 ret = pthread_create(&td->thread, NULL,
1984                                                         thread_main, td);
1985                                 if (ret) {
1986                                         log_err("pthread_create: %s\n",
1987                                                         strerror(ret));
1988                                         nr_started--;
1989                                         break;
1990                                 }
1991                                 ret = pthread_detach(td->thread);
1992                                 if (ret)
1993                                         log_err("pthread_detach: %s",
1994                                                         strerror(ret));
1995                         } else {
1996                                 pid_t pid;
1997                                 dprint(FD_PROCESS, "will fork\n");
1998                                 pid = fork();
1999                                 if (!pid) {
2000                                         int ret = fork_main(shm_id, i);
2001
2002                                         _exit(ret);
2003                                 } else if (i == fio_debug_jobno)
2004                                         *fio_debug_jobp = pid;
2005                         }
2006                         dprint(FD_MUTEX, "wait on startup_mutex\n");
2007                         if (fio_mutex_down_timeout(startup_mutex, 10)) {
2008                                 log_err("fio: job startup hung? exiting.\n");
2009                                 fio_terminate_threads(TERMINATE_ALL);
2010                                 fio_abort = 1;
2011                                 nr_started--;
2012                                 break;
2013                         }
2014                         dprint(FD_MUTEX, "done waiting on startup_mutex\n");
2015                 }
2016
2017                 /*
2018                  * Wait for the started threads to transition to
2019                  * TD_INITIALIZED.
2020                  */
2021                 fio_gettime(&this_start, NULL);
2022                 left = this_jobs;
2023                 while (left && !fio_abort) {
2024                         if (mtime_since_now(&this_start) > JOB_START_TIMEOUT)
2025                                 break;
2026
2027                         do_usleep(100000);
2028
2029                         for (i = 0; i < this_jobs; i++) {
2030                                 td = map[i];
2031                                 if (!td)
2032                                         continue;
2033                                 if (td->runstate == TD_INITIALIZED) {
2034                                         map[i] = NULL;
2035                                         left--;
2036                                 } else if (td->runstate >= TD_EXITED) {
2037                                         map[i] = NULL;
2038                                         left--;
2039                                         todo--;
2040                                         nr_running++; /* work-around... */
2041                                 }
2042                         }
2043                 }
2044
2045                 if (left) {
2046                         log_err("fio: %d job%s failed to start\n", left,
2047                                         left > 1 ? "s" : "");
2048                         for (i = 0; i < this_jobs; i++) {
2049                                 td = map[i];
2050                                 if (!td)
2051                                         continue;
2052                                 kill(td->pid, SIGTERM);
2053                         }
2054                         break;
2055                 }
2056
2057                 /*
2058                  * start created threads (TD_INITIALIZED -> TD_RUNNING).
2059                  */
2060                 for_each_td(td, i) {
2061                         if (td->runstate != TD_INITIALIZED)
2062                                 continue;
2063
2064                         if (in_ramp_time(td))
2065                                 td_set_runstate(td, TD_RAMP);
2066                         else
2067                                 td_set_runstate(td, TD_RUNNING);
2068                         nr_running++;
2069                         nr_started--;
2070                         m_rate += ddir_rw_sum(td->o.ratemin);
2071                         t_rate += ddir_rw_sum(td->o.rate);
2072                         todo--;
2073                         fio_mutex_up(td->mutex);
2074                 }
2075
2076                 reap_threads(&nr_running, &t_rate, &m_rate);
2077
2078                 if (todo)
2079                         do_usleep(100000);
2080         }
2081
2082         while (nr_running) {
2083                 reap_threads(&nr_running, &t_rate, &m_rate);
2084                 do_usleep(10000);
2085         }
2086
2087         fio_idle_prof_stop();
2088
2089         update_io_ticks();
2090 }
2091
2092 static void wait_for_helper_thread_exit(void)
2093 {
2094         void *ret;
2095
2096         helper_exit = 1;
2097         pthread_cond_signal(&helper_cond);
2098         pthread_join(helper_thread, &ret);
2099 }
2100
2101 static void free_disk_util(void)
2102 {
2103         disk_util_prune_entries();
2104
2105         pthread_cond_destroy(&helper_cond);
2106 }
2107
2108 static void *helper_thread_main(void *data)
2109 {
2110         int ret = 0;
2111
2112         fio_mutex_up(startup_mutex);
2113
2114         while (!ret) {
2115                 uint64_t sec = DISK_UTIL_MSEC / 1000;
2116                 uint64_t nsec = (DISK_UTIL_MSEC % 1000) * 1000000;
2117                 struct timespec ts;
2118                 struct timeval tv;
2119
2120                 gettimeofday(&tv, NULL);
2121                 ts.tv_sec = tv.tv_sec + sec;
2122                 ts.tv_nsec = (tv.tv_usec * 1000) + nsec;
2123                 if (ts.tv_nsec > 1000000000ULL) {
2124                         ts.tv_nsec -= 1000000000ULL;
2125                         ts.tv_sec++;
2126                 }
2127
2128                 pthread_cond_timedwait(&helper_cond, &helper_lock, &ts);
2129
2130                 ret = update_io_ticks();
2131
2132                 if (helper_do_stat) {
2133                         helper_do_stat = 0;
2134                         __show_running_run_stats();
2135                 }
2136
2137                 if (!is_backend)
2138                         print_thread_status();
2139         }
2140
2141         return NULL;
2142 }
2143
2144 static int create_helper_thread(void)
2145 {
2146         int ret;
2147
2148         setup_disk_util();
2149
2150         pthread_cond_init(&helper_cond, NULL);
2151         pthread_mutex_init(&helper_lock, NULL);
2152
2153         ret = pthread_create(&helper_thread, NULL, helper_thread_main, NULL);
2154         if (ret) {
2155                 log_err("Can't create helper thread: %s\n", strerror(ret));
2156                 return 1;
2157         }
2158
2159         dprint(FD_MUTEX, "wait on startup_mutex\n");
2160         fio_mutex_down(startup_mutex);
2161         dprint(FD_MUTEX, "done waiting on startup_mutex\n");
2162         return 0;
2163 }
2164
2165 int fio_backend(void)
2166 {
2167         struct thread_data *td;
2168         int i;
2169
2170         if (exec_profile) {
2171                 if (load_profile(exec_profile))
2172                         return 1;
2173                 free(exec_profile);
2174                 exec_profile = NULL;
2175         }
2176         if (!thread_number)
2177                 return 0;
2178
2179         if (write_bw_log) {
2180                 struct log_params p = {
2181                         .log_type = IO_LOG_TYPE_BW,
2182                 };
2183
2184                 setup_log(&agg_io_log[DDIR_READ], &p, "agg-read_bw.log");
2185                 setup_log(&agg_io_log[DDIR_WRITE], &p, "agg-write_bw.log");
2186                 setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log");
2187         }
2188
2189         startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED);
2190         if (startup_mutex == NULL)
2191                 return 1;
2192
2193         set_genesis_time();
2194         stat_init();
2195         create_helper_thread();
2196
2197         cgroup_list = smalloc(sizeof(*cgroup_list));
2198         INIT_FLIST_HEAD(cgroup_list);
2199
2200         run_threads();
2201
2202         wait_for_helper_thread_exit();
2203
2204         if (!fio_abort) {
2205                 __show_run_stats();
2206                 if (write_bw_log) {
2207                         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
2208                                 struct io_log *log = agg_io_log[i];
2209
2210                                 flush_log(log);
2211                                 free_log(log);
2212                         }
2213                 }
2214         }
2215
2216         for_each_td(td, i) {
2217                 fio_options_free(td);
2218                 if (td->rusage_sem) {
2219                         fio_mutex_remove(td->rusage_sem);
2220                         td->rusage_sem = NULL;
2221                 }
2222         }
2223
2224         free_disk_util();
2225         cgroup_kill(cgroup_list);
2226         sfree(cgroup_list);
2227         sfree(cgroup_mnt);
2228
2229         fio_mutex_remove(startup_mutex);
2230         stat_exit();
2231         return exit_value;
2232 }