iolog.c

   1 /*
   2  * Code related to writing an iolog of what a thread is doing, and to
   3  * later read that back and replay
   4  */
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <libgen.h>
   8 #include <assert.h>
   9 #include "flist.h"
  10 #include "fio.h"
  11 #include "verify.h"
  12 #include "trim.h"
  13 #include "filelock.h"
  14
  15 static const char iolog_ver2[] = "fio version 2 iolog";
  16
  17 void queue_io_piece(struct thread_data *td, struct io_piece *ipo)
  18 {
  19         flist_add_tail(&ipo->list, &td->io_log_list);
  20         td->total_io_size += ipo->len;
  21 }
  22
  23 void log_io_u(struct thread_data *td, struct io_u *io_u)
  24 {
  25         const char *act[] = { "read", "write", "sync", "datasync",
  26                                 "sync_file_range", "wait", "trim" };
  27
  28         assert(io_u->ddir <= 6);
  29
  30         if (!td->o.write_iolog_file)
  31                 return;
  32
  33         fprintf(td->iolog_f, "%s %s %llu %lu\n", io_u->file->file_name,
  34                                                 act[io_u->ddir], io_u->offset,
  35                                                 io_u->buflen);
  36 }
  37
  38 void log_file(struct thread_data *td, struct fio_file *f,
  39               enum file_log_act what)
  40 {
  41         const char *act[] = { "add", "open", "close" };
  42
  43         assert(what < 3);
  44
  45         if (!td->o.write_iolog_file)
  46                 return;
  47
  48
  49         /*
  50          * this happens on the pre-open/close done before the job starts
  51          */
  52         if (!td->iolog_f)
  53                 return;
  54
  55         fprintf(td->iolog_f, "%s %s\n", f->file_name, act[what]);
  56 }
  57
  58 static void iolog_delay(struct thread_data *td, unsigned long delay)
  59 {
  60         unsigned long usec = utime_since_now(&td->last_issue);
  61
  62         if (delay < usec)
  63                 return;
  64
  65         delay -= usec;
  66
  67         /*
  68          * less than 100 usec delay, just regard it as noise
  69          */
  70         if (delay < 100)
  71                 return;
  72
  73         usec_sleep(td, delay);
  74 }
  75
  76 static int ipo_special(struct thread_data *td, struct io_piece *ipo)
  77 {
  78         struct fio_file *f;
  79         int ret;
  80
  81         /*
  82          * Not a special ipo
  83          */
  84         if (ipo->ddir != DDIR_INVAL)
  85                 return 0;
  86
  87         f = td->files[ipo->fileno];
  88
  89         switch (ipo->file_action) {
  90         case FIO_LOG_OPEN_FILE:
  91                 ret = td_io_open_file(td, f);
  92                 if (!ret)
  93                         break;
  94                 td_verror(td, ret, "iolog open file");
  95                 return -1;
  96         case FIO_LOG_CLOSE_FILE:
  97                 td_io_close_file(td, f);
  98                 break;
  99         case FIO_LOG_UNLINK_FILE:
 100                 unlink(f->file_name);
 101                 break;
 102         default:
 103                 log_err("fio: bad file action %d\n", ipo->file_action);
 104                 break;
 105         }
 106
 107         return 1;
 108 }
 109
 110 int read_iolog_get(struct thread_data *td, struct io_u *io_u)
 111 {
 112         struct io_piece *ipo;
 113         unsigned long elapsed;
 114
 115         while (!flist_empty(&td->io_log_list)) {
 116                 int ret;
 117
 118                 ipo = flist_entry(td->io_log_list.next, struct io_piece, list);
 119                 flist_del(&ipo->list);
 120                 remove_trim_entry(td, ipo);
 121
 122                 ret = ipo_special(td, ipo);
 123                 if (ret < 0) {
 124                         free(ipo);
 125                         break;
 126                 } else if (ret > 0) {
 127                         free(ipo);
 128                         continue;
 129                 }
 130
 131                 io_u->ddir = ipo->ddir;
 132                 if (ipo->ddir != DDIR_WAIT) {
 133                         io_u->offset = ipo->offset;
 134                         io_u->buflen = ipo->len;
 135                         io_u->file = td->files[ipo->fileno];
 136                         get_file(io_u->file);
 137                         dprint(FD_IO, "iolog: get %llu/%lu/%s\n", io_u->offset,
 138                                                 io_u->buflen, io_u->file->file_name);
 139                         if (ipo->delay)
 140                                 iolog_delay(td, ipo->delay);
 141                 } else {
 142                         elapsed = mtime_since_genesis();
 143                         if (ipo->delay > elapsed)
 144                                 usec_sleep(td, (ipo->delay - elapsed) * 1000);
 145                 }
 146
 147                 free(ipo);
 148
 149                 if (io_u->ddir != DDIR_WAIT)
 150                         return 0;
 151         }
 152
 153         td->done = 1;
 154         return 1;
 155 }
 156
 157 void prune_io_piece_log(struct thread_data *td)
 158 {
 159         struct io_piece *ipo;
 160         struct rb_node *n;
 161
 162         while ((n = rb_first(&td->io_hist_tree)) != NULL) {
 163                 ipo = rb_entry(n, struct io_piece, rb_node);
 164                 rb_erase(n, &td->io_hist_tree);
 165                 remove_trim_entry(td, ipo);
 166                 td->io_hist_len--;
 167                 free(ipo);
 168         }
 169
 170         while (!flist_empty(&td->io_hist_list)) {
 171                 ipo = flist_entry(td->io_hist_list.next, struct io_piece, list);
 172                 flist_del(&ipo->list);
 173                 remove_trim_entry(td, ipo);
 174                 td->io_hist_len--;
 175                 free(ipo);
 176         }
 177 }
 178
 179 /*
 180  * log a successful write, so we can unwind the log for verify
 181  */
 182 void log_io_piece(struct thread_data *td, struct io_u *io_u)
 183 {
 184         struct rb_node **p, *parent;
 185         struct io_piece *ipo, *__ipo;
 186
 187         ipo = malloc(sizeof(struct io_piece));
 188         init_ipo(ipo);
 189         ipo->file = io_u->file;
 190         ipo->offset = io_u->offset;
 191         ipo->len = io_u->buflen;
 192         ipo->numberio = io_u->numberio;
 193         ipo->flags = IP_F_IN_FLIGHT;
 194
 195         io_u->ipo = ipo;
 196
 197         if (io_u_should_trim(td, io_u)) {
 198                 flist_add_tail(&ipo->trim_list, &td->trim_list);
 199                 td->trim_entries++;
 200         }
 201
 202         /*
 203          * We don't need to sort the entries, if:
 204          *
 205          *      Sequential writes, or
 206          *      Random writes that lay out the file as it goes along
 207          *
 208          * For both these cases, just reading back data in the order we
 209          * wrote it out is the fastest.
 210          *
 211          * One exception is if we don't have a random map AND we are doing
 212          * verifies, in that case we need to check for duplicate blocks and
 213          * drop the old one, which we rely on the rb insert/lookup for
 214          * handling.
 215          */
 216         if (((!td->o.verifysort) || !td_random(td) || !td->o.overwrite) &&
 217               (file_randommap(td, ipo->file) || td->o.verify == VERIFY_NONE)) {
 218                 INIT_FLIST_HEAD(&ipo->list);
 219                 flist_add_tail(&ipo->list, &td->io_hist_list);
 220                 ipo->flags |= IP_F_ONLIST;
 221                 td->io_hist_len++;
 222                 return;
 223         }
 224
 225         RB_CLEAR_NODE(&ipo->rb_node);
 226
 227         /*
 228          * Sort the entry into the verification list
 229          */
 230 restart:
 231         p = &td->io_hist_tree.rb_node;
 232         parent = NULL;
 233         while (*p) {
 234                 parent = *p;
 235
 236                 __ipo = rb_entry(parent, struct io_piece, rb_node);
 237                 if (ipo->file < __ipo->file)
 238                         p = &(*p)->rb_left;
 239                 else if (ipo->file > __ipo->file)
 240                         p = &(*p)->rb_right;
 241                 else if (ipo->offset < __ipo->offset)
 242                         p = &(*p)->rb_left;
 243                 else if (ipo->offset > __ipo->offset)
 244                         p = &(*p)->rb_right;
 245                 else {
 246                         dprint(FD_IO, "iolog: overlap %llu/%lu, %llu/%lu",
 247                                 __ipo->offset, __ipo->len,
 248                                 ipo->offset, ipo->len);
 249                         td->io_hist_len--;
 250                         rb_erase(parent, &td->io_hist_tree);
 251                         remove_trim_entry(td, __ipo);
 252                         free(__ipo);
 253                         goto restart;
 254                 }
 255         }
 256
 257         rb_link_node(&ipo->rb_node, parent, p);
 258         rb_insert_color(&ipo->rb_node, &td->io_hist_tree);
 259         ipo->flags |= IP_F_ONRB;
 260         td->io_hist_len++;
 261 }
 262
 263 void write_iolog_close(struct thread_data *td)
 264 {
 265         fflush(td->iolog_f);
 266         fclose(td->iolog_f);
 267         free(td->iolog_buf);
 268         td->iolog_f = NULL;
 269         td->iolog_buf = NULL;
 270 }
 271
 272 /*
 273  * Read version 2 iolog data. It is enhanced to include per-file logging,
 274  * syncs, etc.
 275  */
 276 static int read_iolog2(struct thread_data *td, FILE *f)
 277 {
 278         unsigned long long offset;
 279         unsigned int bytes;
 280         int reads, writes, waits, fileno = 0, file_action = 0; /* stupid gcc */
 281         char *fname, *act;
 282         char *str, *p;
 283         enum fio_ddir rw;
 284
 285         free_release_files(td);
 286
 287         /*
 288          * Read in the read iolog and store it, reuse the infrastructure
 289          * for doing verifications.
 290          */
 291         str = malloc(4096);
 292         fname = malloc(256+16);
 293         act = malloc(256+16);
 294
 295         reads = writes = waits = 0;
 296         while ((p = fgets(str, 4096, f)) != NULL) {
 297                 struct io_piece *ipo;
 298                 int r;
 299
 300                 r = sscanf(p, "%256s %256s %llu %u", fname, act, &offset,
 301                                                                         &bytes);
 302                 if (r == 4) {
 303                         /*
 304                          * Check action first
 305                          */
 306                         if (!strcmp(act, "wait"))
 307                                 rw = DDIR_WAIT;
 308                         else if (!strcmp(act, "read"))
 309                                 rw = DDIR_READ;
 310                         else if (!strcmp(act, "write"))
 311                                 rw = DDIR_WRITE;
 312                         else if (!strcmp(act, "sync"))
 313                                 rw = DDIR_SYNC;
 314                         else if (!strcmp(act, "datasync"))
 315                                 rw = DDIR_DATASYNC;
 316                         else if (!strcmp(act, "trim"))
 317                                 rw = DDIR_TRIM;
 318                         else {
 319                                 log_err("fio: bad iolog file action: %s\n",
 320                                                                         act);
 321                                 continue;
 322                         }
 323                         fileno = get_fileno(td, fname);
 324                 } else if (r == 2) {
 325                         rw = DDIR_INVAL;
 326                         if (!strcmp(act, "add")) {
 327                                 fileno = add_file(td, fname, 0, 1);
 328                                 file_action = FIO_LOG_ADD_FILE;
 329                                 continue;
 330                         } else if (!strcmp(act, "open")) {
 331                                 fileno = get_fileno(td, fname);
 332                                 file_action = FIO_LOG_OPEN_FILE;
 333                         } else if (!strcmp(act, "close")) {
 334                                 fileno = get_fileno(td, fname);
 335                                 file_action = FIO_LOG_CLOSE_FILE;
 336                         } else {
 337                                 log_err("fio: bad iolog file action: %s\n",
 338                                                                         act);
 339                                 continue;
 340                         }
 341                 } else {
 342                         log_err("bad iolog2: %s", p);
 343                         continue;
 344                 }
 345
 346                 if (rw == DDIR_READ)
 347                         reads++;
 348                 else if (rw == DDIR_WRITE) {
 349                         /*
 350                          * Don't add a write for ro mode
 351                          */
 352                         if (read_only)
 353                                 continue;
 354                         writes++;
 355                 } else if (rw == DDIR_WAIT) {
 356                         waits++;
 357                 } else if (rw == DDIR_INVAL) {
 358                 } else if (!ddir_sync(rw)) {
 359                         log_err("bad ddir: %d\n", rw);
 360                         continue;
 361                 }
 362
 363                 /*
 364                  * Make note of file
 365                  */
 366                 ipo = malloc(sizeof(*ipo));
 367                 init_ipo(ipo);
 368                 ipo->ddir = rw;
 369                 if (rw == DDIR_WAIT) {
 370                         ipo->delay = offset;
 371                 } else {
 372                         ipo->offset = offset;
 373                         ipo->len = bytes;
 374                         if (bytes > td->o.max_bs[rw])
 375                                 td->o.max_bs[rw] = bytes;
 376                         ipo->fileno = fileno;
 377                         ipo->file_action = file_action;
 378                 }
 379
 380                 queue_io_piece(td, ipo);
 381         }
 382
 383         free(str);
 384         free(act);
 385         free(fname);
 386
 387         if (writes && read_only) {
 388                 log_err("fio: <%s> skips replay of %d writes due to"
 389                         " read-only\n", td->o.name, writes);
 390                 writes = 0;
 391         }
 392
 393         if (!reads && !writes && !waits)
 394                 return 1;
 395         else if (reads && !writes)
 396                 td->o.td_ddir = TD_DDIR_READ;
 397         else if (!reads && writes)
 398                 td->o.td_ddir = TD_DDIR_WRITE;
 399         else
 400                 td->o.td_ddir = TD_DDIR_RW;
 401
 402         return 0;
 403 }
 404
 405 /*
 406  * open iolog, check version, and call appropriate parser
 407  */
 408 static int init_iolog_read(struct thread_data *td)
 409 {
 410         char buffer[256], *p;
 411         FILE *f;
 412         int ret;
 413
 414         f = fopen(td->o.read_iolog_file, "r");
 415         if (!f) {
 416                 perror("fopen read iolog");
 417                 return 1;
 418         }
 419
 420         p = fgets(buffer, sizeof(buffer), f);
 421         if (!p) {
 422                 td_verror(td, errno, "iolog read");
 423                 log_err("fio: unable to read iolog\n");
 424                 fclose(f);
 425                 return 1;
 426         }
 427
 428         /*
 429          * version 2 of the iolog stores a specific string as the
 430          * first line, check for that
 431          */
 432         if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2)))
 433                 ret = read_iolog2(td, f);
 434         else {
 435                 log_err("fio: iolog version 1 is no longer supported\n");
 436                 ret = 1;
 437         }
 438
 439         fclose(f);
 440         return ret;
 441 }
 442
 443 /*
 444  * Set up a log for storing io patterns.
 445  */
 446 static int init_iolog_write(struct thread_data *td)
 447 {
 448         struct fio_file *ff;
 449         FILE *f;
 450         unsigned int i;
 451
 452         f = fopen(td->o.write_iolog_file, "a");
 453         if (!f) {
 454                 perror("fopen write iolog");
 455                 return 1;
 456         }
 457
 458         /*
 459          * That's it for writing, setup a log buffer and we're done.
 460           */
 461         td->iolog_f = f;
 462         td->iolog_buf = malloc(8192);
 463         setvbuf(f, td->iolog_buf, _IOFBF, 8192);
 464
 465         /*
 466          * write our version line
 467          */
 468         if (fprintf(f, "%s\n", iolog_ver2) < 0) {
 469                 perror("iolog init\n");
 470                 return 1;
 471         }
 472
 473         /*
 474          * add all known files
 475          */
 476         for_each_file(td, ff, i)
 477                 log_file(td, ff, FIO_LOG_ADD_FILE);
 478
 479         return 0;
 480 }
 481
 482 int init_iolog(struct thread_data *td)
 483 {
 484         int ret = 0;
 485
 486         if (td->o.read_iolog_file) {
 487                 int need_swap;
 488
 489                 /*
 490                  * Check if it's a blktrace file and load that if possible.
 491                  * Otherwise assume it's a normal log file and load that.
 492                  */
 493                 if (is_blktrace(td->o.read_iolog_file, &need_swap))
 494                         ret = load_blktrace(td, td->o.read_iolog_file, need_swap);
 495                 else
 496                         ret = init_iolog_read(td);
 497         } else if (td->o.write_iolog_file)
 498                 ret = init_iolog_write(td);
 499
 500         if (ret)
 501                 td_verror(td, EINVAL, "failed initializing iolog");
 502
 503         return ret;
 504 }
 505
 506 void setup_log(struct io_log **log, unsigned long avg_msec, int log_type)
 507 {
 508         struct io_log *l = malloc(sizeof(*l));
 509
 510         memset(l, 0, sizeof(*l));
 511         l->nr_samples = 0;
 512         l->max_samples = 1024;
 513         l->log_type = log_type;
 514         l->log = malloc(l->max_samples * sizeof(struct io_sample));
 515         l->avg_msec = avg_msec;
 516         *log = l;
 517 }
 518
 519 #ifdef CONFIG_SETVBUF
 520 static void *set_file_buffer(FILE *f)
 521 {
 522         size_t size = 1048576;
 523         void *buf;
 524
 525         buf = malloc(size);
 526         setvbuf(f, buf, _IOFBF, size);
 527         return buf;
 528 }
 529
 530 static void clear_file_buffer(void *buf)
 531 {
 532         free(buf);
 533 }
 534 #else
 535 static void *set_file_buffer(FILE *f)
 536 {
 537         return NULL;
 538 }
 539
 540 static void clear_file_buffer(void *buf)
 541 {
 542 }
 543 #endif
 544
 545 void __finish_log(struct io_log *log, const char *name)
 546 {
 547         unsigned int i;
 548         void *buf;
 549         FILE *f;
 550
 551         f = fopen(name, "a");
 552         if (!f) {
 553                 perror("fopen log");
 554                 return;
 555         }
 556
 557         buf = set_file_buffer(f);
 558
 559         for (i = 0; i < log->nr_samples; i++) {
 560                 fprintf(f, "%lu, %lu, %u, %u\n",
 561                                 (unsigned long) log->log[i].time,
 562                                 (unsigned long) log->log[i].val,
 563                                 log->log[i].ddir, log->log[i].bs);
 564         }
 565
 566         fclose(f);
 567         clear_file_buffer(buf);
 568         free(log->log);
 569         free(log);
 570 }
 571
 572 static int finish_log_named(struct thread_data *td, struct io_log *log,
 573                             const char *prefix, const char *postfix,
 574                             int trylock)
 575 {
 576         char file_name[256];
 577
 578         snprintf(file_name, sizeof(file_name), "%s_%s.log", prefix, postfix);
 579
 580         if (trylock) {
 581                 if (fio_trylock_file(file_name))
 582                         return 1;
 583         } else
 584                 fio_lock_file(file_name);
 585
 586         if (td->client_type == FIO_CLIENT_TYPE_GUI) {
 587                 fio_send_iolog(td, log, file_name);
 588                 free(log->log);
 589                 free(log);
 590         } else
 591                 __finish_log(log, file_name);
 592
 593         fio_unlock_file(file_name);
 594         return 0;
 595 }
 596
 597 static int finish_log(struct thread_data *td, struct io_log *log,
 598                       const char *name, int trylock)
 599 {
 600         return finish_log_named(td, log, td->o.name, name, trylock);
 601 }
 602
 603 static int write_this_log(struct thread_data *td, struct io_log *log,
 604                           const char *log_file, const char *name, int try)
 605 {
 606         int ret;
 607
 608         if (!log)
 609                 return 0;
 610
 611         if (log_file)
 612                 ret = finish_log_named(td, log, log_file, name, try);
 613         else
 614                 ret = finish_log(td, log, name, try);
 615
 616         return ret;
 617 }
 618
 619 static int write_iops_log(struct thread_data *td, int try)
 620 {
 621         struct thread_options *o = &td->o;
 622
 623         return write_this_log(td, td->iops_log, o->iops_log_file, "iops", try);
 624 }
 625
 626 static int write_slat_log(struct thread_data *td, int try)
 627 {
 628         struct thread_options *o = &td->o;
 629
 630         return write_this_log(td, td->slat_log, o->lat_log_file, "slat", try);
 631 }
 632
 633 static int write_clat_log(struct thread_data *td, int try)
 634 {
 635         struct thread_options *o = &td->o;
 636
 637         return write_this_log(td, td->clat_log, o->lat_log_file, "clat" , try);
 638 }
 639
 640 static int write_lat_log(struct thread_data *td, int try)
 641 {
 642         struct thread_options *o = &td->o;
 643
 644         return write_this_log(td, td->lat_log, o->lat_log_file, "lat", try);
 645 }
 646
 647 static int write_bandw_log(struct thread_data *td, int try)
 648 {
 649         struct thread_options *o = &td->o;
 650
 651         return write_this_log(td, td->bw_log, o->bw_log_file, "bw", try);
 652 }
 653
 654 enum {
 655         BW_LOG_MASK     = 1,
 656         LAT_LOG_MASK    = 2,
 657         SLAT_LOG_MASK   = 4,
 658         CLAT_LOG_MASK   = 8,
 659         IOPS_LOG_MASK   = 16,
 660
 661         ALL_LOG_NR      = 5,
 662 };
 663
 664 struct log_type {
 665         unsigned int mask;
 666         int (*fn)(struct thread_data *, int);
 667 };
 668
 669 static struct log_type log_types[] = {
 670         {
 671                 .mask   = BW_LOG_MASK,
 672                 .fn     = write_bandw_log,
 673         },
 674         {
 675                 .mask   = LAT_LOG_MASK,
 676                 .fn     = write_lat_log,
 677         },
 678         {
 679                 .mask   = SLAT_LOG_MASK,
 680                 .fn     = write_slat_log,
 681         },
 682         {
 683                 .mask   = CLAT_LOG_MASK,
 684                 .fn     = write_clat_log,
 685         },
 686         {
 687                 .mask   = IOPS_LOG_MASK,
 688                 .fn     = write_iops_log,
 689         },
 690 };
 691
 692 void fio_writeout_logs(struct thread_data *td)
 693 {
 694         unsigned int log_mask = 0;
 695         unsigned int log_left = ALL_LOG_NR;
 696         int old_state, i;
 697
 698         old_state = td_bump_runstate(td, TD_FINISHING);
 699
 700         finalize_logs(td);
 701
 702         while (log_left) {
 703                 int prev_log_left = log_left;
 704
 705                 for (i = 0; i < ALL_LOG_NR && log_left; i++) {
 706                         struct log_type *lt = &log_types[i];
 707                         int ret;
 708
 709                         if (!(log_mask & lt->mask)) {
 710                                 ret = lt->fn(td, log_left != 1);
 711                                 if (!ret) {
 712                                         log_left--;
 713                                         log_mask |= lt->mask;
 714                                 }
 715                         }
 716                 }
 717
 718                 if (prev_log_left == log_left)
 719                         usleep(5000);
 720         }
 721
 722         td_restore_runstate(td, old_state);
 723 }