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