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