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