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 <sys/types.h>
  10 #include <sys/stat.h>
  11 #include <unistd.h>
  12 #ifdef CONFIG_ZLIB
  13 #include <zlib.h>
  14 #endif
  15
  16 #include "flist.h"
  17 #include "fio.h"
  18 #include "verify.h"
  19 #include "trim.h"
  20 #include "filelock.h"
  21 #include "lib/tp.h"
  22
  23 static const char iolog_ver2[] = "fio version 2 iolog";
  24
  25 void queue_io_piece(struct thread_data *td, struct io_piece *ipo)
  26 {
  27         flist_add_tail(&ipo->list, &td->io_log_list);
  28         td->total_io_size += ipo->len;
  29 }
  30
  31 void log_io_u(const struct thread_data *td, const struct io_u *io_u)
  32 {
  33         if (!td->o.write_iolog_file)
  34                 return;
  35
  36         fprintf(td->iolog_f, "%s %s %llu %lu\n", io_u->file->file_name,
  37                                                 io_ddir_name(io_u->ddir),
  38                                                 io_u->offset, io_u->buflen);
  39 }
  40
  41 void log_file(struct thread_data *td, struct fio_file *f,
  42               enum file_log_act what)
  43 {
  44         const char *act[] = { "add", "open", "close" };
  45
  46         assert(what < 3);
  47
  48         if (!td->o.write_iolog_file)
  49                 return;
  50
  51
  52         /*
  53          * this happens on the pre-open/close done before the job starts
  54          */
  55         if (!td->iolog_f)
  56                 return;
  57
  58         fprintf(td->iolog_f, "%s %s\n", f->file_name, act[what]);
  59 }
  60
  61 static void iolog_delay(struct thread_data *td, unsigned long delay)
  62 {
  63         uint64_t usec = utime_since_now(&td->last_issue);
  64         uint64_t this_delay;
  65         struct timeval tv;
  66
  67         if (delay < td->time_offset) {
  68                 td->time_offset = 0;
  69                 return;
  70         }
  71
  72         delay -= td->time_offset;
  73         if (delay < usec)
  74                 return;
  75
  76         delay -= usec;
  77
  78         fio_gettime(&tv, NULL);
  79         while (delay && !td->terminate) {
  80                 this_delay = delay;
  81                 if (this_delay > 500000)
  82                         this_delay = 500000;
  83
  84                 usec_sleep(td, this_delay);
  85                 delay -= this_delay;
  86         }
  87
  88         usec = utime_since_now(&tv);
  89         if (usec > delay)
  90                 td->time_offset = usec - delay;
  91         else
  92                 td->time_offset = 0;
  93 }
  94
  95 static int ipo_special(struct thread_data *td, struct io_piece *ipo)
  96 {
  97         struct fio_file *f;
  98         int ret;
  99
 100         /*
 101          * Not a special ipo
 102          */
 103         if (ipo->ddir != DDIR_INVAL)
 104                 return 0;
 105
 106         f = td->files[ipo->fileno];
 107
 108         switch (ipo->file_action) {
 109         case FIO_LOG_OPEN_FILE:
 110                 ret = td_io_open_file(td, f);
 111                 if (!ret)
 112                         break;
 113                 td_verror(td, ret, "iolog open file");
 114                 return -1;
 115         case FIO_LOG_CLOSE_FILE:
 116                 td_io_close_file(td, f);
 117                 break;
 118         case FIO_LOG_UNLINK_FILE:
 119                 td_io_unlink_file(td, f);
 120                 break;
 121         default:
 122                 log_err("fio: bad file action %d\n", ipo->file_action);
 123                 break;
 124         }
 125
 126         return 1;
 127 }
 128
 129 int read_iolog_get(struct thread_data *td, struct io_u *io_u)
 130 {
 131         struct io_piece *ipo;
 132         unsigned long elapsed;
 133
 134         while (!flist_empty(&td->io_log_list)) {
 135                 int ret;
 136
 137                 ipo = flist_first_entry(&td->io_log_list, struct io_piece, list);
 138                 flist_del(&ipo->list);
 139                 remove_trim_entry(td, ipo);
 140
 141                 ret = ipo_special(td, ipo);
 142                 if (ret < 0) {
 143                         free(ipo);
 144                         break;
 145                 } else if (ret > 0) {
 146                         free(ipo);
 147                         continue;
 148                 }
 149
 150                 io_u->ddir = ipo->ddir;
 151                 if (ipo->ddir != DDIR_WAIT) {
 152                         io_u->offset = ipo->offset;
 153                         io_u->buflen = ipo->len;
 154                         io_u->file = td->files[ipo->fileno];
 155                         get_file(io_u->file);
 156                         dprint(FD_IO, "iolog: get %llu/%lu/%s\n", io_u->offset,
 157                                                 io_u->buflen, io_u->file->file_name);
 158                         if (ipo->delay)
 159                                 iolog_delay(td, ipo->delay);
 160                 } else {
 161                         elapsed = mtime_since_genesis();
 162                         if (ipo->delay > elapsed)
 163                                 usec_sleep(td, (ipo->delay - elapsed) * 1000);
 164                 }
 165
 166                 free(ipo);
 167
 168                 if (io_u->ddir != DDIR_WAIT)
 169                         return 0;
 170         }
 171
 172         td->done = 1;
 173         return 1;
 174 }
 175
 176 void prune_io_piece_log(struct thread_data *td)
 177 {
 178         struct io_piece *ipo;
 179         struct rb_node *n;
 180
 181         while ((n = rb_first(&td->io_hist_tree)) != NULL) {
 182                 ipo = rb_entry(n, struct io_piece, rb_node);
 183                 rb_erase(n, &td->io_hist_tree);
 184                 remove_trim_entry(td, ipo);
 185                 td->io_hist_len--;
 186                 free(ipo);
 187         }
 188
 189         while (!flist_empty(&td->io_hist_list)) {
 190                 ipo = flist_first_entry(&td->io_hist_list, struct io_piece, list);
 191                 flist_del(&ipo->list);
 192                 remove_trim_entry(td, ipo);
 193                 td->io_hist_len--;
 194                 free(ipo);
 195         }
 196 }
 197
 198 /*
 199  * log a successful write, so we can unwind the log for verify
 200  */
 201 void log_io_piece(struct thread_data *td, struct io_u *io_u)
 202 {
 203         struct rb_node **p, *parent;
 204         struct io_piece *ipo, *__ipo;
 205
 206         ipo = malloc(sizeof(struct io_piece));
 207         init_ipo(ipo);
 208         ipo->file = io_u->file;
 209         ipo->offset = io_u->offset;
 210         ipo->len = io_u->buflen;
 211         ipo->numberio = io_u->numberio;
 212         ipo->flags = IP_F_IN_FLIGHT;
 213
 214         io_u->ipo = ipo;
 215
 216         if (io_u_should_trim(td, io_u)) {
 217                 flist_add_tail(&ipo->trim_list, &td->trim_list);
 218                 td->trim_entries++;
 219         }
 220
 221         /*
 222          * We don't need to sort the entries, if:
 223          *
 224          *      Sequential writes, or
 225          *      Random writes that lay out the file as it goes along
 226          *
 227          * For both these cases, just reading back data in the order we
 228          * wrote it out is the fastest.
 229          *
 230          * One exception is if we don't have a random map AND we are doing
 231          * verifies, in that case we need to check for duplicate blocks and
 232          * drop the old one, which we rely on the rb insert/lookup for
 233          * handling.
 234          */
 235         if (((!td->o.verifysort) || !td_random(td) || !td->o.overwrite) &&
 236               (file_randommap(td, ipo->file) || td->o.verify == VERIFY_NONE)) {
 237                 INIT_FLIST_HEAD(&ipo->list);
 238                 flist_add_tail(&ipo->list, &td->io_hist_list);
 239                 ipo->flags |= IP_F_ONLIST;
 240                 td->io_hist_len++;
 241                 return;
 242         }
 243
 244         RB_CLEAR_NODE(&ipo->rb_node);
 245
 246         /*
 247          * Sort the entry into the verification list
 248          */
 249 restart:
 250         p = &td->io_hist_tree.rb_node;
 251         parent = NULL;
 252         while (*p) {
 253                 int overlap = 0;
 254                 parent = *p;
 255
 256                 __ipo = rb_entry(parent, struct io_piece, rb_node);
 257                 if (ipo->file < __ipo->file)
 258                         p = &(*p)->rb_left;
 259                 else if (ipo->file > __ipo->file)
 260                         p = &(*p)->rb_right;
 261                 else if (ipo->offset < __ipo->offset) {
 262                         p = &(*p)->rb_left;
 263                         overlap = ipo->offset + ipo->len > __ipo->offset;
 264                 }
 265                 else if (ipo->offset > __ipo->offset) {
 266                         p = &(*p)->rb_right;
 267                         overlap = __ipo->offset + __ipo->len > ipo->offset;
 268                 }
 269                 else
 270                         overlap = 1;
 271
 272                 if (overlap) {
 273                         dprint(FD_IO, "iolog: overlap %llu/%lu, %llu/%lu",
 274                                 __ipo->offset, __ipo->len,
 275                                 ipo->offset, ipo->len);
 276                         td->io_hist_len--;
 277                         rb_erase(parent, &td->io_hist_tree);
 278                         remove_trim_entry(td, __ipo);
 279                         free(__ipo);
 280                         goto restart;
 281                 }
 282         }
 283
 284         rb_link_node(&ipo->rb_node, parent, p);
 285         rb_insert_color(&ipo->rb_node, &td->io_hist_tree);
 286         ipo->flags |= IP_F_ONRB;
 287         td->io_hist_len++;
 288 }
 289
 290 void unlog_io_piece(struct thread_data *td, struct io_u *io_u)
 291 {
 292         struct io_piece *ipo = io_u->ipo;
 293
 294         if (!ipo)
 295                 return;
 296
 297         if (ipo->flags & IP_F_ONRB)
 298                 rb_erase(&ipo->rb_node, &td->io_hist_tree);
 299         else if (ipo->flags & IP_F_ONLIST)
 300                 flist_del(&ipo->list);
 301
 302         free(ipo);
 303         io_u->ipo = NULL;
 304         td->io_hist_len--;
 305 }
 306
 307 void trim_io_piece(struct thread_data *td, const struct io_u *io_u)
 308 {
 309         struct io_piece *ipo = io_u->ipo;
 310
 311         if (!ipo)
 312                 return;
 313
 314         ipo->len = io_u->xfer_buflen - io_u->resid;
 315 }
 316
 317 void write_iolog_close(struct thread_data *td)
 318 {
 319         fflush(td->iolog_f);
 320         fclose(td->iolog_f);
 321         free(td->iolog_buf);
 322         td->iolog_f = NULL;
 323         td->iolog_buf = NULL;
 324 }
 325
 326 /*
 327  * Read version 2 iolog data. It is enhanced to include per-file logging,
 328  * syncs, etc.
 329  */
 330 static int read_iolog2(struct thread_data *td, FILE *f)
 331 {
 332         unsigned long long offset;
 333         unsigned int bytes;
 334         int reads, writes, waits, fileno = 0, file_action = 0; /* stupid gcc */
 335         char *fname, *act;
 336         char *str, *p;
 337         enum fio_ddir rw;
 338
 339         free_release_files(td);
 340
 341         /*
 342          * Read in the read iolog and store it, reuse the infrastructure
 343          * for doing verifications.
 344          */
 345         str = malloc(4096);
 346         fname = malloc(256+16);
 347         act = malloc(256+16);
 348
 349         reads = writes = waits = 0;
 350         while ((p = fgets(str, 4096, f)) != NULL) {
 351                 struct io_piece *ipo;
 352                 int r;
 353
 354                 r = sscanf(p, "%256s %256s %llu %u", fname, act, &offset,
 355                                                                         &bytes);
 356                 if (r == 4) {
 357                         /*
 358                          * Check action first
 359                          */
 360                         if (!strcmp(act, "wait"))
 361                                 rw = DDIR_WAIT;
 362                         else if (!strcmp(act, "read"))
 363                                 rw = DDIR_READ;
 364                         else if (!strcmp(act, "write"))
 365                                 rw = DDIR_WRITE;
 366                         else if (!strcmp(act, "sync"))
 367                                 rw = DDIR_SYNC;
 368                         else if (!strcmp(act, "datasync"))
 369                                 rw = DDIR_DATASYNC;
 370                         else if (!strcmp(act, "trim"))
 371                                 rw = DDIR_TRIM;
 372                         else {
 373                                 log_err("fio: bad iolog file action: %s\n",
 374                                                                         act);
 375                                 continue;
 376                         }
 377                         fileno = get_fileno(td, fname);
 378                 } else if (r == 2) {
 379                         rw = DDIR_INVAL;
 380                         if (!strcmp(act, "add")) {
 381                                 fileno = add_file(td, fname, 0, 1);
 382                                 file_action = FIO_LOG_ADD_FILE;
 383                                 continue;
 384                         } else if (!strcmp(act, "open")) {
 385                                 fileno = get_fileno(td, fname);
 386                                 file_action = FIO_LOG_OPEN_FILE;
 387                         } else if (!strcmp(act, "close")) {
 388                                 fileno = get_fileno(td, fname);
 389                                 file_action = FIO_LOG_CLOSE_FILE;
 390                         } else {
 391                                 log_err("fio: bad iolog file action: %s\n",
 392                                                                         act);
 393                                 continue;
 394                         }
 395                 } else {
 396                         log_err("bad iolog2: %s", p);
 397                         continue;
 398                 }
 399
 400                 if (rw == DDIR_READ)
 401                         reads++;
 402                 else if (rw == DDIR_WRITE) {
 403                         /*
 404                          * Don't add a write for ro mode
 405                          */
 406                         if (read_only)
 407                                 continue;
 408                         writes++;
 409                 } else if (rw == DDIR_WAIT) {
 410                         waits++;
 411                 } else if (rw == DDIR_INVAL) {
 412                 } else if (!ddir_sync(rw)) {
 413                         log_err("bad ddir: %d\n", rw);
 414                         continue;
 415                 }
 416
 417                 /*
 418                  * Make note of file
 419                  */
 420                 ipo = malloc(sizeof(*ipo));
 421                 init_ipo(ipo);
 422                 ipo->ddir = rw;
 423                 if (rw == DDIR_WAIT) {
 424                         ipo->delay = offset;
 425                 } else {
 426                         ipo->offset = offset;
 427                         ipo->len = bytes;
 428                         if (rw != DDIR_INVAL && bytes > td->o.max_bs[rw])
 429                                 td->o.max_bs[rw] = bytes;
 430                         ipo->fileno = fileno;
 431                         ipo->file_action = file_action;
 432                         td->o.size += bytes;
 433                 }
 434
 435                 queue_io_piece(td, ipo);
 436         }
 437
 438         free(str);
 439         free(act);
 440         free(fname);
 441
 442         if (writes && read_only) {
 443                 log_err("fio: <%s> skips replay of %d writes due to"
 444                         " read-only\n", td->o.name, writes);
 445                 writes = 0;
 446         }
 447
 448         if (!reads && !writes && !waits)
 449                 return 1;
 450         else if (reads && !writes)
 451                 td->o.td_ddir = TD_DDIR_READ;
 452         else if (!reads && writes)
 453                 td->o.td_ddir = TD_DDIR_WRITE;
 454         else
 455                 td->o.td_ddir = TD_DDIR_RW;
 456
 457         return 0;
 458 }
 459
 460 /*
 461  * open iolog, check version, and call appropriate parser
 462  */
 463 static int init_iolog_read(struct thread_data *td)
 464 {
 465         char buffer[256], *p;
 466         FILE *f;
 467         int ret;
 468
 469         f = fopen(td->o.read_iolog_file, "r");
 470         if (!f) {
 471                 perror("fopen read iolog");
 472                 return 1;
 473         }
 474
 475         p = fgets(buffer, sizeof(buffer), f);
 476         if (!p) {
 477                 td_verror(td, errno, "iolog read");
 478                 log_err("fio: unable to read iolog\n");
 479                 fclose(f);
 480                 return 1;
 481         }
 482
 483         /*
 484          * version 2 of the iolog stores a specific string as the
 485          * first line, check for that
 486          */
 487         if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2)))
 488                 ret = read_iolog2(td, f);
 489         else {
 490                 log_err("fio: iolog version 1 is no longer supported\n");
 491                 ret = 1;
 492         }
 493
 494         fclose(f);
 495         return ret;
 496 }
 497
 498 /*
 499  * Set up a log for storing io patterns.
 500  */
 501 static int init_iolog_write(struct thread_data *td)
 502 {
 503         struct fio_file *ff;
 504         FILE *f;
 505         unsigned int i;
 506
 507         f = fopen(td->o.write_iolog_file, "a");
 508         if (!f) {
 509                 perror("fopen write iolog");
 510                 return 1;
 511         }
 512
 513         /*
 514          * That's it for writing, setup a log buffer and we're done.
 515           */
 516         td->iolog_f = f;
 517         td->iolog_buf = malloc(8192);
 518         setvbuf(f, td->iolog_buf, _IOFBF, 8192);
 519
 520         /*
 521          * write our version line
 522          */
 523         if (fprintf(f, "%s\n", iolog_ver2) < 0) {
 524                 perror("iolog init\n");
 525                 return 1;
 526         }
 527
 528         /*
 529          * add all known files
 530          */
 531         for_each_file(td, ff, i)
 532                 log_file(td, ff, FIO_LOG_ADD_FILE);
 533
 534         return 0;
 535 }
 536
 537 int init_iolog(struct thread_data *td)
 538 {
 539         int ret = 0;
 540
 541         if (td->o.read_iolog_file) {
 542                 int need_swap;
 543
 544                 /*
 545                  * Check if it's a blktrace file and load that if possible.
 546                  * Otherwise assume it's a normal log file and load that.
 547                  */
 548                 if (is_blktrace(td->o.read_iolog_file, &need_swap))
 549                         ret = load_blktrace(td, td->o.read_iolog_file, need_swap);
 550                 else
 551                         ret = init_iolog_read(td);
 552         } else if (td->o.write_iolog_file)
 553                 ret = init_iolog_write(td);
 554
 555         if (ret)
 556                 td_verror(td, EINVAL, "failed initializing iolog");
 557
 558         return ret;
 559 }
 560
 561 void setup_log(struct io_log **log, struct log_params *p,
 562                const char *filename)
 563 {
 564         struct io_log *l;
 565
 566         l = calloc(1, sizeof(*l));
 567         l->nr_samples = 0;
 568         l->max_samples = 1024;
 569         l->log_type = p->log_type;
 570         l->log_offset = p->log_offset;
 571         l->log_gz = p->log_gz;
 572         l->log_gz_store = p->log_gz_store;
 573         l->log = malloc(l->max_samples * log_entry_sz(l));
 574         l->avg_msec = p->avg_msec;
 575         l->filename = strdup(filename);
 576         l->td = p->td;
 577
 578         if (l->log_offset)
 579                 l->log_ddir_mask = LOG_OFFSET_SAMPLE_BIT;
 580
 581         INIT_FLIST_HEAD(&l->chunk_list);
 582
 583         if (l->log_gz && !p->td)
 584                 l->log_gz = 0;
 585         else if (l->log_gz) {
 586                 pthread_mutex_init(&l->chunk_lock, NULL);
 587                 p->td->flags |= TD_F_COMPRESS_LOG;
 588         }
 589
 590         *log = l;
 591 }
 592
 593 #ifdef CONFIG_SETVBUF
 594 static void *set_file_buffer(FILE *f)
 595 {
 596         size_t size = 1048576;
 597         void *buf;
 598
 599         buf = malloc(size);
 600         setvbuf(f, buf, _IOFBF, size);
 601         return buf;
 602 }
 603
 604 static void clear_file_buffer(void *buf)
 605 {
 606         free(buf);
 607 }
 608 #else
 609 static void *set_file_buffer(FILE *f)
 610 {
 611         return NULL;
 612 }
 613
 614 static void clear_file_buffer(void *buf)
 615 {
 616 }
 617 #endif
 618
 619 void free_log(struct io_log *log)
 620 {
 621         free(log->log);
 622         free(log->filename);
 623         free(log);
 624 }
 625
 626 static void flush_samples(FILE *f, void *samples, uint64_t sample_size)
 627 {
 628         struct io_sample *s;
 629         int log_offset;
 630         uint64_t i, nr_samples;
 631
 632         if (!sample_size)
 633                 return;
 634
 635         s = __get_sample(samples, 0, 0);
 636         log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
 637
 638         nr_samples = sample_size / __log_entry_sz(log_offset);
 639
 640         for (i = 0; i < nr_samples; i++) {
 641                 s = __get_sample(samples, log_offset, i);
 642
 643                 if (!log_offset) {
 644                         fprintf(f, "%lu, %lu, %u, %u\n",
 645                                         (unsigned long) s->time,
 646                                         (unsigned long) s->val,
 647                                         io_sample_ddir(s), s->bs);
 648                 } else {
 649                         struct io_sample_offset *so = (void *) s;
 650
 651                         fprintf(f, "%lu, %lu, %u, %u, %llu\n",
 652                                         (unsigned long) s->time,
 653                                         (unsigned long) s->val,
 654                                         io_sample_ddir(s), s->bs,
 655                                         (unsigned long long) so->offset);
 656                 }
 657         }
 658 }
 659
 660 #ifdef CONFIG_ZLIB
 661
 662 struct iolog_flush_data {
 663         struct tp_work work;
 664         struct io_log *log;
 665         void *samples;
 666         uint64_t nr_samples;
 667 };
 668
 669 struct iolog_compress {
 670         struct flist_head list;
 671         void *buf;
 672         size_t len;
 673         unsigned int seq;
 674 };
 675
 676 #define GZ_CHUNK        131072
 677
 678 static struct iolog_compress *get_new_chunk(unsigned int seq)
 679 {
 680         struct iolog_compress *c;
 681
 682         c = malloc(sizeof(*c));
 683         INIT_FLIST_HEAD(&c->list);
 684         c->buf = malloc(GZ_CHUNK);
 685         c->len = 0;
 686         c->seq = seq;
 687         return c;
 688 }
 689
 690 static void free_chunk(struct iolog_compress *ic)
 691 {
 692         free(ic->buf);
 693         free(ic);
 694 }
 695
 696 static int z_stream_init(z_stream *stream, int gz_hdr)
 697 {
 698         int wbits = 15;
 699
 700         stream->zalloc = Z_NULL;
 701         stream->zfree = Z_NULL;
 702         stream->opaque = Z_NULL;
 703         stream->next_in = Z_NULL;
 704
 705         /*
 706          * zlib magic - add 32 for auto-detection of gz header or not,
 707          * if we decide to store files in a gzip friendly format.
 708          */
 709         if (gz_hdr)
 710                 wbits += 32;
 711
 712         if (inflateInit2(stream, wbits) != Z_OK)
 713                 return 1;
 714
 715         return 0;
 716 }
 717
 718 struct inflate_chunk_iter {
 719         unsigned int seq;
 720         int err;
 721         void *buf;
 722         size_t buf_size;
 723         size_t buf_used;
 724         size_t chunk_sz;
 725 };
 726
 727 static void finish_chunk(z_stream *stream, FILE *f,
 728                          struct inflate_chunk_iter *iter)
 729 {
 730         int ret;
 731
 732         ret = inflateEnd(stream);
 733         if (ret != Z_OK)
 734                 log_err("fio: failed to end log inflation (%d)\n", ret);
 735
 736         flush_samples(f, iter->buf, iter->buf_used);
 737         free(iter->buf);
 738         iter->buf = NULL;
 739         iter->buf_size = iter->buf_used = 0;
 740 }
 741
 742 /*
 743  * Iterative chunk inflation. Handles cases where we cross into a new
 744  * sequence, doing flush finish of previous chunk if needed.
 745  */
 746 static size_t inflate_chunk(struct iolog_compress *ic, int gz_hdr, FILE *f,
 747                             z_stream *stream, struct inflate_chunk_iter *iter)
 748 {
 749         size_t ret;
 750
 751         dprint(FD_COMPRESS, "inflate chunk size=%lu, seq=%u",
 752                                 (unsigned long) ic->len, ic->seq);
 753
 754         if (ic->seq != iter->seq) {
 755                 if (iter->seq)
 756                         finish_chunk(stream, f, iter);
 757
 758                 z_stream_init(stream, gz_hdr);
 759                 iter->seq = ic->seq;
 760         }
 761
 762         stream->avail_in = ic->len;
 763         stream->next_in = ic->buf;
 764
 765         if (!iter->buf_size) {
 766                 iter->buf_size = iter->chunk_sz;
 767                 iter->buf = malloc(iter->buf_size);
 768         }
 769
 770         while (stream->avail_in) {
 771                 size_t this_out = iter->buf_size - iter->buf_used;
 772                 int err;
 773
 774                 stream->avail_out = this_out;
 775                 stream->next_out = iter->buf + iter->buf_used;
 776
 777                 err = inflate(stream, Z_NO_FLUSH);
 778                 if (err < 0) {
 779                         log_err("fio: failed inflating log: %d\n", err);
 780                         iter->err = err;
 781                         break;
 782                 }
 783
 784                 iter->buf_used += this_out - stream->avail_out;
 785
 786                 if (!stream->avail_out) {
 787                         iter->buf_size += iter->chunk_sz;
 788                         iter->buf = realloc(iter->buf, iter->buf_size);
 789                         continue;
 790                 }
 791
 792                 if (err == Z_STREAM_END)
 793                         break;
 794         }
 795
 796         ret = (void *) stream->next_in - ic->buf;
 797
 798         dprint(FD_COMPRESS, "inflated to size=%lu\n", (unsigned long) ret);
 799
 800         return ret;
 801 }
 802
 803 /*
 804  * Inflate stored compressed chunks, or write them directly to the log
 805  * file if so instructed.
 806  */
 807 static int inflate_gz_chunks(struct io_log *log, FILE *f)
 808 {
 809         struct inflate_chunk_iter iter = { .chunk_sz = log->log_gz, };
 810         z_stream stream;
 811
 812         while (!flist_empty(&log->chunk_list)) {
 813                 struct iolog_compress *ic;
 814
 815                 ic = flist_first_entry(&log->chunk_list, struct iolog_compress, list);
 816                 flist_del(&ic->list);
 817
 818                 if (log->log_gz_store) {
 819                         size_t ret;
 820
 821                         dprint(FD_COMPRESS, "log write chunk size=%lu, "
 822                                 "seq=%u\n", (unsigned long) ic->len, ic->seq);
 823
 824                         ret = fwrite(ic->buf, ic->len, 1, f);
 825                         if (ret != 1 || ferror(f)) {
 826                                 iter.err = errno;
 827                                 log_err("fio: error writing compressed log\n");
 828                         }
 829                 } else
 830                         inflate_chunk(ic, log->log_gz_store, f, &stream, &iter);
 831
 832                 free_chunk(ic);
 833         }
 834
 835         if (iter.seq) {
 836                 finish_chunk(&stream, f, &iter);
 837                 free(iter.buf);
 838         }
 839
 840         return iter.err;
 841 }
 842
 843 /*
 844  * Open compressed log file and decompress the stored chunks and
 845  * write them to stdout. The chunks are stored sequentially in the
 846  * file, so we iterate over them and do them one-by-one.
 847  */
 848 int iolog_file_inflate(const char *file)
 849 {
 850         struct inflate_chunk_iter iter = { .chunk_sz = 64 * 1024 * 1024, };
 851         struct iolog_compress ic;
 852         z_stream stream;
 853         struct stat sb;
 854         ssize_t ret;
 855         size_t total;
 856         void *buf;
 857         FILE *f;
 858
 859         f = fopen(file, "r");
 860         if (!f) {
 861                 perror("fopen");
 862                 return 1;
 863         }
 864
 865         if (stat(file, &sb) < 0) {
 866                 fclose(f);
 867                 perror("stat");
 868                 return 1;
 869         }
 870
 871         ic.buf = buf = malloc(sb.st_size);
 872         ic.len = sb.st_size;
 873         ic.seq = 1;
 874
 875         ret = fread(ic.buf, ic.len, 1, f);
 876         if (ret < 0) {
 877                 perror("fread");
 878                 fclose(f);
 879                 free(buf);
 880                 return 1;
 881         } else if (ret != 1) {
 882                 log_err("fio: short read on reading log\n");
 883                 fclose(f);
 884                 free(buf);
 885                 return 1;
 886         }
 887
 888         fclose(f);
 889
 890         /*
 891          * Each chunk will return Z_STREAM_END. We don't know how many
 892          * chunks are in the file, so we just keep looping and incrementing
 893          * the sequence number until we have consumed the whole compressed
 894          * file.
 895          */
 896         total = ic.len;
 897         do {
 898                 size_t iret;
 899
 900                 iret = inflate_chunk(&ic,  1, stdout, &stream, &iter);
 901                 total -= iret;
 902                 if (!total)
 903                         break;
 904                 if (iter.err)
 905                         break;
 906
 907                 ic.seq++;
 908                 ic.len -= iret;
 909                 ic.buf += iret;
 910         } while (1);
 911
 912         if (iter.seq) {
 913                 finish_chunk(&stream, stdout, &iter);
 914                 free(iter.buf);
 915         }
 916
 917         free(buf);
 918         return iter.err;
 919 }
 920
 921 #else
 922
 923 static int inflate_gz_chunks(struct io_log *log, FILE *f)
 924 {
 925         return 0;
 926 }
 927
 928 int iolog_file_inflate(const char *file)
 929 {
 930         log_err("fio: log inflation not possible without zlib\n");
 931         return 1;
 932 }
 933
 934 #endif
 935
 936 void flush_log(struct io_log *log)
 937 {
 938         void *buf;
 939         FILE *f;
 940
 941         f = fopen(log->filename, "w");
 942         if (!f) {
 943                 perror("fopen log");
 944                 return;
 945         }
 946
 947         buf = set_file_buffer(f);
 948
 949         inflate_gz_chunks(log, f);
 950
 951         flush_samples(f, log->log, log->nr_samples * log_entry_sz(log));
 952
 953         fclose(f);
 954         clear_file_buffer(buf);
 955 }
 956
 957 static int finish_log(struct thread_data *td, struct io_log *log, int trylock)
 958 {
 959         if (td->tp_data)
 960                 iolog_flush(log, 1);
 961
 962         if (trylock) {
 963                 if (fio_trylock_file(log->filename))
 964                         return 1;
 965         } else
 966                 fio_lock_file(log->filename);
 967
 968         if (td->client_type == FIO_CLIENT_TYPE_GUI)
 969                 fio_send_iolog(td, log, log->filename);
 970         else
 971                 flush_log(log);
 972
 973         fio_unlock_file(log->filename);
 974         free_log(log);
 975         return 0;
 976 }
 977
 978 #ifdef CONFIG_ZLIB
 979
 980 /*
 981  * Invoked from our compress helper thread, when logging would have exceeded
 982  * the specified memory limitation. Compresses the previously stored
 983  * entries.
 984  */
 985 static int gz_work(struct tp_work *work)
 986 {
 987         struct iolog_flush_data *data;
 988         struct iolog_compress *c;
 989         struct flist_head list;
 990         unsigned int seq;
 991         z_stream stream;
 992         size_t total = 0;
 993         int ret;
 994
 995         INIT_FLIST_HEAD(&list);
 996
 997         data = container_of(work, struct iolog_flush_data, work);
 998
 999         stream.zalloc = Z_NULL;
1000         stream.zfree = Z_NULL;
1001         stream.opaque = Z_NULL;
1002
1003         ret = deflateInit(&stream, Z_DEFAULT_COMPRESSION);
1004         if (ret != Z_OK) {
1005                 log_err("fio: failed to init gz stream\n");
1006                 return 0;
1007         }
1008
1009         seq = ++data->log->chunk_seq;
1010
1011         stream.next_in = (void *) data->samples;
1012         stream.avail_in = data->nr_samples * log_entry_sz(data->log);
1013
1014         dprint(FD_COMPRESS, "deflate input size=%lu, seq=%u\n",
1015                                 (unsigned long) stream.avail_in, seq);
1016         do {
1017                 c = get_new_chunk(seq);
1018                 stream.avail_out = GZ_CHUNK;
1019                 stream.next_out = c->buf;
1020                 ret = deflate(&stream, Z_NO_FLUSH);
1021                 if (ret < 0) {
1022                         log_err("fio: deflate log (%d)\n", ret);
1023                         free_chunk(c);
1024                         goto err;
1025                 }
1026
1027                 c->len = GZ_CHUNK - stream.avail_out;
1028                 flist_add_tail(&c->list, &list);
1029                 total += c->len;
1030         } while (stream.avail_in);
1031
1032         stream.next_out = c->buf + c->len;
1033         stream.avail_out = GZ_CHUNK - c->len;
1034
1035         ret = deflate(&stream, Z_FINISH);
1036         if (ret == Z_STREAM_END)
1037                 c->len = GZ_CHUNK - stream.avail_out;
1038         else {
1039                 do {
1040                         c = get_new_chunk(seq);
1041                         stream.avail_out = GZ_CHUNK;
1042                         stream.next_out = c->buf;
1043                         ret = deflate(&stream, Z_FINISH);
1044                         c->len = GZ_CHUNK - stream.avail_out;
1045                         total += c->len;
1046                         flist_add_tail(&c->list, &list);
1047                 } while (ret != Z_STREAM_END);
1048         }
1049
1050         dprint(FD_COMPRESS, "deflated to size=%lu\n", (unsigned long) total);
1051
1052         ret = deflateEnd(&stream);
1053         if (ret != Z_OK)
1054                 log_err("fio: deflateEnd %d\n", ret);
1055
1056         free(data->samples);
1057
1058         if (!flist_empty(&list)) {
1059                 pthread_mutex_lock(&data->log->chunk_lock);
1060                 flist_splice_tail(&list, &data->log->chunk_list);
1061                 pthread_mutex_unlock(&data->log->chunk_lock);
1062         }
1063
1064         ret = 0;
1065 done:
1066         if (work->wait) {
1067                 work->done = 1;
1068                 pthread_cond_signal(&work->cv);
1069         } else
1070                 free(data);
1071
1072         return ret;
1073 err:
1074         while (!flist_empty(&list)) {
1075                 c = flist_first_entry(list.next, struct iolog_compress, list);
1076                 flist_del(&c->list);
1077                 free_chunk(c);
1078         }
1079         ret = 1;
1080         goto done;
1081 }
1082
1083 /*
1084  * Queue work item to compress the existing log entries. We copy the
1085  * samples, and reset the log sample count to 0 (so the logging will
1086  * continue to use the memory associated with the log). If called with
1087  * wait == 1, will not return until the log compression has completed.
1088  */
1089 int iolog_flush(struct io_log *log, int wait)
1090 {
1091         struct tp_data *tdat = log->td->tp_data;
1092         struct iolog_flush_data *data;
1093         size_t sample_size;
1094
1095         data = malloc(sizeof(*data));
1096         if (!data)
1097                 return 1;
1098
1099         data->log = log;
1100
1101         sample_size = log->nr_samples * log_entry_sz(log);
1102         data->samples = malloc(sample_size);
1103         if (!data->samples) {
1104                 free(data);
1105                 return 1;
1106         }
1107
1108         memcpy(data->samples, log->log, sample_size);
1109         data->nr_samples = log->nr_samples;
1110         data->work.fn = gz_work;
1111         log->nr_samples = 0;
1112
1113         if (wait) {
1114                 pthread_mutex_init(&data->work.lock, NULL);
1115                 pthread_cond_init(&data->work.cv, NULL);
1116                 data->work.wait = 1;
1117         } else
1118                 data->work.wait = 0;
1119
1120         data->work.prio = 1;
1121         tp_queue_work(tdat, &data->work);
1122
1123         if (wait) {
1124                 pthread_mutex_lock(&data->work.lock);
1125                 while (!data->work.done)
1126                         pthread_cond_wait(&data->work.cv, &data->work.lock);
1127                 pthread_mutex_unlock(&data->work.lock);
1128                 free(data);
1129         }
1130
1131         return 0;
1132 }
1133
1134 #else
1135
1136 int iolog_flush(struct io_log *log, int wait)
1137 {
1138         return 1;
1139 }
1140
1141 #endif
1142
1143 static int write_iops_log(struct thread_data *td, int try)
1144 {
1145         struct io_log *log = td->iops_log;
1146
1147         if (!log)
1148                 return 0;
1149
1150         return finish_log(td, log, try);
1151 }
1152
1153 static int write_slat_log(struct thread_data *td, int try)
1154 {
1155         struct io_log *log = td->slat_log;
1156
1157         if (!log)
1158                 return 0;
1159
1160         return finish_log(td, log, try);
1161 }
1162
1163 static int write_clat_log(struct thread_data *td, int try)
1164 {
1165         struct io_log *log = td->clat_log;
1166
1167         if (!log)
1168                 return 0;
1169
1170         return finish_log(td, log, try);
1171 }
1172
1173 static int write_lat_log(struct thread_data *td, int try)
1174 {
1175         struct io_log *log = td->lat_log;
1176
1177         if (!log)
1178                 return 0;
1179
1180         return finish_log(td, log, try);
1181 }
1182
1183 static int write_bandw_log(struct thread_data *td, int try)
1184 {
1185         struct io_log *log = td->bw_log;
1186
1187         if (!log)
1188                 return 0;
1189
1190         return finish_log(td, log, try);
1191 }
1192
1193 enum {
1194         BW_LOG_MASK     = 1,
1195         LAT_LOG_MASK    = 2,
1196         SLAT_LOG_MASK   = 4,
1197         CLAT_LOG_MASK   = 8,
1198         IOPS_LOG_MASK   = 16,
1199
1200         ALL_LOG_NR      = 5,
1201 };
1202
1203 struct log_type {
1204         unsigned int mask;
1205         int (*fn)(struct thread_data *, int);
1206 };
1207
1208 static struct log_type log_types[] = {
1209         {
1210                 .mask   = BW_LOG_MASK,
1211                 .fn     = write_bandw_log,
1212         },
1213         {
1214                 .mask   = LAT_LOG_MASK,
1215                 .fn     = write_lat_log,
1216         },
1217         {
1218                 .mask   = SLAT_LOG_MASK,
1219                 .fn     = write_slat_log,
1220         },
1221         {
1222                 .mask   = CLAT_LOG_MASK,
1223                 .fn     = write_clat_log,
1224         },
1225         {
1226                 .mask   = IOPS_LOG_MASK,
1227                 .fn     = write_iops_log,
1228         },
1229 };
1230
1231 void fio_writeout_logs(struct thread_data *td)
1232 {
1233         unsigned int log_mask = 0;
1234         unsigned int log_left = ALL_LOG_NR;
1235         int old_state, i;
1236
1237         old_state = td_bump_runstate(td, TD_FINISHING);
1238
1239         finalize_logs(td);
1240
1241         while (log_left) {
1242                 int prev_log_left = log_left;
1243
1244                 for (i = 0; i < ALL_LOG_NR && log_left; i++) {
1245                         struct log_type *lt = &log_types[i];
1246                         int ret;
1247
1248                         if (!(log_mask & lt->mask)) {
1249                                 ret = lt->fn(td, log_left != 1);
1250                                 if (!ret) {
1251                                         log_left--;
1252                                         log_mask |= lt->mask;
1253                                 }
1254                         }
1255                 }
1256
1257                 if (prev_log_left == log_left)
1258                         usleep(5000);
1259         }
1260
1261         td_restore_runstate(td, old_state);
1262 }