ltp/fsx.c

   1 /*
   2  *      Copyright (C) 1991, NeXT Computer, Inc.  All Rights Reserverd.
   3  *
   4  *      File:   fsx.c
   5  *      Author: Avadis Tevanian, Jr.
   6  *
   7  *      File system exerciser.
   8  *
   9  *      Rewritten 8/98 by Conrad Minshall.
  10  *
  11  *      Small changes to work under Linux -- davej.
  12  *
  13  *      Checks for mmap last-page zero fill.
  14  */
  15
  16 #include "global.h"
  17
  18 #include <limits.h>
  19 #include <time.h>
  20 #include <strings.h>
  21 #include <sys/file.h>
  22 #include <sys/mman.h>
  23 #ifdef HAVE_ERR_H
  24 #include <err.h>
  25 #endif
  26 #include <signal.h>
  27 #include <stdio.h>
  28 #include <stdlib.h>
  29 #include <string.h>
  30 #include <stdarg.h>
  31 #include <errno.h>
  32 #ifdef AIO
  33 #include <libaio.h>
  34 #endif
  35
  36 #ifndef MAP_FILE
  37 # define MAP_FILE 0
  38 #endif
  39
  40 #define NUMPRINTCOLUMNS 32      /* # columns of data to print on each line */
  41
  42 /*
  43  *      A log entry is an operation and a bunch of arguments.
  44  */
  45
  46 struct log_entry {
  47         int     operation;
  48         int     args[3];
  49 };
  50
  51 #define LOGSIZE 1000
  52
  53 struct log_entry        oplog[LOGSIZE]; /* the log */
  54 int                     logptr = 0;     /* current position in log */
  55 int                     logcount = 0;   /* total ops */
  56
  57 /*
  58  *      Define operations
  59  */
  60
  61 #define OP_READ         1
  62 #define OP_WRITE        2
  63 #define OP_TRUNCATE     3
  64 #define OP_CLOSEOPEN    4
  65 #define OP_MAPREAD      5
  66 #define OP_MAPWRITE     6
  67 #define OP_SKIPPED      7
  68
  69 #undef PAGE_SIZE
  70 #define PAGE_SIZE       getpagesize()
  71 #undef PAGE_MASK
  72 #define PAGE_MASK       (PAGE_SIZE - 1)
  73
  74 char    *original_buf;                  /* a pointer to the original data */
  75 char    *good_buf;                      /* a pointer to the correct data */
  76 char    *temp_buf;                      /* a pointer to the current data */
  77 char    *fname;                         /* name of our test file */
  78 int     fd;                             /* fd for our test file */
  79
  80 off_t           file_size = 0;
  81 off_t           biggest = 0;
  82 char            state[256];
  83 unsigned long   testcalls = 0;          /* calls to function "test" */
  84
  85 unsigned long   simulatedopcount = 0;   /* -b flag */
  86 int     closeprob = 0;                  /* -c flag */
  87 int     debug = 0;                      /* -d flag */
  88 unsigned long   debugstart = 0;         /* -D flag */
  89 int     flush = 0;                      /* -f flag */
  90 int     do_fsync = 0;                   /* -y flag */
  91 unsigned long   maxfilelen = 256 * 1024;        /* -l flag */
  92 int     sizechecks = 1;                 /* -n flag disables them */
  93 int     maxoplen = 64 * 1024;           /* -o flag */
  94 int     quiet = 0;                      /* -q flag */
  95 unsigned long progressinterval = 0;     /* -p flag */
  96 int     readbdy = 1;                    /* -r flag */
  97 int     style = 0;                      /* -s flag */
  98 int     prealloc = 0;                   /* -x flag */
  99 int     truncbdy = 1;                   /* -t flag */
 100 int     writebdy = 1;                   /* -w flag */
 101 long    monitorstart = -1;              /* -m flag */
 102 long    monitorend = -1;                /* -m flag */
 103 int     lite = 0;                       /* -L flag */
 104 long    numops = -1;                    /* -N flag */
 105 int     randomoplen = 1;                /* -O flag disables it */
 106 int     seed = 1;                       /* -S flag */
 107 int     mapped_writes = 1;              /* -W flag disables */
 108 int     mapped_reads = 1;               /* -R flag disables it */
 109 int     fsxgoodfd = 0;
 110 int     o_direct;                       /* -Z */
 111 int     aio = 0;
 112
 113 int page_size;
 114 int page_mask;
 115 int mmap_mask;
 116 #ifdef AIO
 117 int aio_rw(int rw, int fd, char *buf, unsigned len, unsigned offset);
 118 #define READ 0
 119 #define WRITE 1
 120 #define fsxread(a,b,c,d)        aio_rw(READ, a,b,c,d)
 121 #define fsxwrite(a,b,c,d)       aio_rw(WRITE, a,b,c,d)
 122 #else
 123 #define fsxread(a,b,c,d)        read(a,b,c)
 124 #define fsxwrite(a,b,c,d)       write(a,b,c)
 125 #endif
 126
 127 FILE *  fsxlogf = NULL;
 128 int badoff = -1;
 129 int closeopen = 0;
 130
 131 static void *round_up(void *ptr, unsigned long align, unsigned long offset)
 132 {
 133         unsigned long ret = (unsigned long)ptr;
 134
 135         ret = ((ret + align - 1) & ~(align - 1));
 136         ret += offset;
 137         return (void *)ret;
 138 }
 139
 140 void
 141 vwarnc(int code, const char *fmt, va_list ap) {
 142   fprintf(stderr, "fsx: ");
 143   if (fmt != NULL) {
 144         vfprintf(stderr, fmt, ap);
 145         fprintf(stderr, ": ");
 146   }
 147   fprintf(stderr, "%s\n", strerror(code));
 148 }
 149
 150 void
 151 warn(const char * fmt, ...)  {
 152         va_list ap;
 153         va_start(ap, fmt);
 154         vwarnc(errno, fmt, ap);
 155         va_end(ap);
 156 }
 157
 158 #define BUF_SIZE 1024
 159
 160 void
 161 prt(char *fmt, ...)
 162 {
 163         va_list args;
 164         char buffer[BUF_SIZE];
 165
 166         va_start(args, fmt);
 167         vsnprintf(buffer, BUF_SIZE, fmt, args);
 168         va_end(args);
 169         fprintf(stdout, buffer);
 170         if (fsxlogf)
 171                 fprintf(fsxlogf, buffer);
 172 }
 173
 174 void
 175 prterr(char *prefix)
 176 {
 177         prt("%s%s%s\n", prefix, prefix ? ": " : "", strerror(errno));
 178 }
 179
 180
 181 void
 182 log4(int operation, int arg0, int arg1, int arg2)
 183 {
 184         struct log_entry *le;
 185
 186         le = &oplog[logptr];
 187         le->operation = operation;
 188         if (closeopen)
 189                 le->operation = ~ le->operation;
 190         le->args[0] = arg0;
 191         le->args[1] = arg1;
 192         le->args[2] = arg2;
 193         logptr++;
 194         logcount++;
 195         if (logptr >= LOGSIZE)
 196                 logptr = 0;
 197 }
 198
 199
 200 void
 201 logdump(void)
 202 {
 203         int     i, count, down;
 204         struct log_entry        *lp;
 205
 206         prt("LOG DUMP (%d total operations):\n", logcount);
 207         if (logcount < LOGSIZE) {
 208                 i = 0;
 209                 count = logcount;
 210         } else {
 211                 i = logptr;
 212                 count = LOGSIZE;
 213         }
 214         for ( ; count > 0; count--) {
 215                 int opnum;
 216
 217                 opnum = i+1 + (logcount/LOGSIZE)*LOGSIZE;
 218                 prt("%d(%d mod 256): ", opnum, opnum%256);
 219                 lp = &oplog[i];
 220                 if ((closeopen = lp->operation < 0))
 221                         lp->operation = ~ lp->operation;
 222
 223                 switch (lp->operation) {
 224                 case OP_MAPREAD:
 225                         prt("MAPREAD\t0x%x thru 0x%x\t(0x%x bytes)",
 226                             lp->args[0], lp->args[0] + lp->args[1] - 1,
 227                             lp->args[1]);
 228                         if (badoff >= lp->args[0] && badoff <
 229                                                      lp->args[0] + lp->args[1])
 230                                 prt("\t***RRRR***");
 231                         break;
 232                 case OP_MAPWRITE:
 233                         prt("MAPWRITE 0x%x thru 0x%x\t(0x%x bytes)",
 234                             lp->args[0], lp->args[0] + lp->args[1] - 1,
 235                             lp->args[1]);
 236                         if (badoff >= lp->args[0] && badoff <
 237                                                      lp->args[0] + lp->args[1])
 238                                 prt("\t******WWWW");
 239                         break;
 240                 case OP_READ:
 241                         prt("READ\t0x%x thru 0x%x\t(0x%x bytes)",
 242                             lp->args[0], lp->args[0] + lp->args[1] - 1,
 243                             lp->args[1]);
 244                         if (badoff >= lp->args[0] &&
 245                             badoff < lp->args[0] + lp->args[1])
 246                                 prt("\t***RRRR***");
 247                         break;
 248                 case OP_WRITE:
 249                         prt("WRITE\t0x%x thru 0x%x\t(0x%x bytes)",
 250                             lp->args[0], lp->args[0] + lp->args[1] - 1,
 251                             lp->args[1]);
 252                         if (lp->args[0] > lp->args[2])
 253                                 prt(" HOLE");
 254                         else if (lp->args[0] + lp->args[1] > lp->args[2])
 255                                 prt(" EXTEND");
 256                         if ((badoff >= lp->args[0] || badoff >=lp->args[2]) &&
 257                             badoff < lp->args[0] + lp->args[1])
 258                                 prt("\t***WWWW");
 259                         break;
 260                 case OP_TRUNCATE:
 261                         down = lp->args[0] < lp->args[1];
 262                         prt("TRUNCATE %s\tfrom 0x%x to 0x%x",
 263                             down ? "DOWN" : "UP", lp->args[1], lp->args[0]);
 264                         if (badoff >= lp->args[!down] &&
 265                             badoff < lp->args[!!down])
 266                                 prt("\t******WWWW");
 267                         break;
 268                 case OP_SKIPPED:
 269                         prt("SKIPPED (no operation)");
 270                         break;
 271                 default:
 272                         prt("BOGUS LOG ENTRY (operation code = %d)!",
 273                             lp->operation);
 274                 }
 275                 if (closeopen)
 276                         prt("\n\t\tCLOSE/OPEN");
 277                 prt("\n");
 278                 i++;
 279                 if (i == LOGSIZE)
 280                         i = 0;
 281         }
 282 }
 283
 284
 285 void
 286 save_buffer(char *buffer, off_t bufferlength, int fd)
 287 {
 288         off_t ret;
 289         ssize_t byteswritten;
 290
 291         if (fd <= 0 || bufferlength == 0)
 292                 return;
 293
 294         if (bufferlength > SSIZE_MAX) {
 295                 prt("fsx flaw: overflow in save_buffer\n");
 296                 exit(67);
 297         }
 298         if (lite) {
 299                 off_t size_by_seek = lseek(fd, (off_t)0, SEEK_END);
 300                 if (size_by_seek == (off_t)-1)
 301                         prterr("save_buffer: lseek eof");
 302                 else if (bufferlength > size_by_seek) {
 303                         warn("save_buffer: .fsxgood file too short... will save 0x%llx bytes instead of 0x%llx\n", (unsigned long long)size_by_seek,
 304                              (unsigned long long)bufferlength);
 305                         bufferlength = size_by_seek;
 306                 }
 307         }
 308
 309         ret = lseek(fd, (off_t)0, SEEK_SET);
 310         if (ret == (off_t)-1)
 311                 prterr("save_buffer: lseek 0");
 312
 313         byteswritten = write(fd, buffer, (size_t)bufferlength);
 314         if (byteswritten != bufferlength) {
 315                 if (byteswritten == -1)
 316                         prterr("save_buffer write");
 317                 else
 318                         warn("save_buffer: short write, 0x%x bytes instead of 0x%llx\n",
 319                              (unsigned)byteswritten,
 320                              (unsigned long long)bufferlength);
 321         }
 322 }
 323
 324
 325 void
 326 report_failure(int status)
 327 {
 328         logdump();
 329
 330         if (fsxgoodfd) {
 331                 if (good_buf) {
 332                         save_buffer(good_buf, file_size, fsxgoodfd);
 333                         prt("Correct content saved for comparison\n");
 334                         prt("(maybe hexdump \"%s\" vs \"%s.fsxgood\")\n",
 335                             fname, fname);
 336                 }
 337                 close(fsxgoodfd);
 338         }
 339         exit(status);
 340 }
 341
 342
 343 #define short_at(cp) ((unsigned short)((*((unsigned char *)(cp)) << 8) | \
 344                                         *(((unsigned char *)(cp)) + 1)))
 345
 346 void
 347 check_buffers(unsigned offset, unsigned size)
 348 {
 349         unsigned char c, t;
 350         unsigned i = 0;
 351         unsigned n = 0;
 352         unsigned op = 0;
 353         unsigned bad = 0;
 354
 355         if (memcmp(good_buf + offset, temp_buf, size) != 0) {
 356                 prt("READ BAD DATA: offset = 0x%x, size = 0x%x, fname = %s\n",
 357                     offset, size, fname);
 358                 prt("OFFSET\tGOOD\tBAD\tRANGE\n");
 359                 while (size > 0) {
 360                         c = good_buf[offset];
 361                         t = temp_buf[i];
 362                         if (c != t) {
 363                                 if (n < 16) {
 364                                         bad = short_at(&temp_buf[i]);
 365                                         prt("0x%5x\t0x%04x\t0x%04x", offset,
 366                                             short_at(&good_buf[offset]), bad);
 367                                         op = temp_buf[offset & 1 ? i+1 : i];
 368                                         prt("\t0x%5x\n", n);
 369                                         if (op)
 370                                                 prt("operation# (mod 256) for "
 371                                                   "the bad data may be %u\n",
 372                                                 ((unsigned)op & 0xff));
 373                                         else
 374                                                 prt("operation# (mod 256) for "
 375                                                   "the bad data unknown, check"
 376                                                   " HOLE and EXTEND ops\n");
 377                                 }
 378                                 n++;
 379                                 badoff = offset;
 380                         }
 381                         offset++;
 382                         i++;
 383                         size--;
 384                 }
 385                 report_failure(110);
 386         }
 387 }
 388
 389
 390 void
 391 check_size(void)
 392 {
 393         struct stat     statbuf;
 394         off_t   size_by_seek;
 395
 396         if (fstat(fd, &statbuf)) {
 397                 prterr("check_size: fstat");
 398                 statbuf.st_size = -1;
 399         }
 400         size_by_seek = lseek(fd, (off_t)0, SEEK_END);
 401         if (file_size != statbuf.st_size || file_size != size_by_seek) {
 402                 prt("Size error: expected 0x%llx stat 0x%llx seek 0x%llx\n",
 403                     (unsigned long long)file_size,
 404                     (unsigned long long)statbuf.st_size,
 405                     (unsigned long long)size_by_seek);
 406                 report_failure(120);
 407         }
 408 }
 409
 410
 411 void
 412 check_trunc_hack(void)
 413 {
 414         struct stat statbuf;
 415
 416         ftruncate(fd, (off_t)0);
 417         ftruncate(fd, (off_t)100000);
 418         fstat(fd, &statbuf);
 419         if (statbuf.st_size != (off_t)100000) {
 420                 prt("no extend on truncate! not posix!\n");
 421                 exit(130);
 422         }
 423         ftruncate(fd, 0);
 424 }
 425
 426 void
 427 doflush(unsigned offset, unsigned size)
 428 {
 429         unsigned pg_offset;
 430         unsigned map_size;
 431         char    *p;
 432
 433         if (o_direct == O_DIRECT)
 434                 return;
 435
 436         pg_offset = offset & mmap_mask;
 437         map_size  = pg_offset + size;
 438
 439         if ((p = (char *)mmap(0, map_size, PROT_READ | PROT_WRITE,
 440                               MAP_FILE | MAP_SHARED, fd,
 441                               (off_t)(offset - pg_offset))) == (char *)-1) {
 442                 prterr("doflush: mmap");
 443                 report_failure(202);
 444         }
 445         if (msync(p, map_size, MS_INVALIDATE) != 0) {
 446                 prterr("doflush: msync");
 447                 report_failure(203);
 448         }
 449         if (munmap(p, map_size) != 0) {
 450                 prterr("doflush: munmap");
 451                 report_failure(204);
 452         }
 453 }
 454
 455 void
 456 doread(unsigned offset, unsigned size)
 457 {
 458         off_t ret;
 459         unsigned iret;
 460
 461         offset -= offset % readbdy;
 462         if (o_direct)
 463                 size -= size % readbdy;
 464         if (size == 0) {
 465                 if (!quiet && testcalls > simulatedopcount && !o_direct)
 466                         prt("skipping zero size read\n");
 467                 log4(OP_SKIPPED, OP_READ, offset, size);
 468                 return;
 469         }
 470         if (size + offset > file_size) {
 471                 if (!quiet && testcalls > simulatedopcount)
 472                         prt("skipping seek/read past end of file\n");
 473                 log4(OP_SKIPPED, OP_READ, offset, size);
 474                 return;
 475         }
 476
 477         log4(OP_READ, offset, size, 0);
 478
 479         if (testcalls <= simulatedopcount)
 480                 return;
 481
 482         if (!quiet &&
 483                 ((progressinterval && testcalls % progressinterval == 0)  ||
 484                 (debug &&
 485                        (monitorstart == -1 ||
 486                         (offset + size > monitorstart &&
 487                         (monitorend == -1 || offset <= monitorend))))))
 488                 prt("%lu read\t0x%x thru\t0x%x\t(0x%x bytes)\n", testcalls,
 489                     offset, offset + size - 1, size);
 490         ret = lseek(fd, (off_t)offset, SEEK_SET);
 491         if (ret == (off_t)-1) {
 492                 prterr("doread: lseek");
 493                 report_failure(140);
 494         }
 495         iret = fsxread(fd, temp_buf, size, offset);
 496         if (iret != size) {
 497                 if (iret == -1)
 498                         prterr("doread: read");
 499                 else
 500                         prt("short read: 0x%x bytes instead of 0x%x\n",
 501                             iret, size);
 502                 report_failure(141);
 503         }
 504         check_buffers(offset, size);
 505 }
 506
 507
 508 void
 509 check_eofpage(char *s, unsigned offset, char *p, int size)
 510 {
 511         unsigned long last_page, should_be_zero;
 512
 513         if (offset + size <= (file_size & ~page_mask))
 514                 return;
 515         /*
 516          * we landed in the last page of the file
 517          * test to make sure the VM system provided 0's
 518          * beyond the true end of the file mapping
 519          * (as required by mmap def in 1996 posix 1003.1)
 520          */
 521         last_page = ((unsigned long)p + (offset & page_mask) + size) & ~page_mask;
 522
 523         for (should_be_zero = last_page + (file_size & page_mask);
 524              should_be_zero < last_page + page_size;
 525              should_be_zero++)
 526                 if (*(char *)should_be_zero) {
 527                         prt("Mapped %s: non-zero data past EOF (0x%llx) page offset 0x%x is 0x%04x\n",
 528                             s, file_size - 1, should_be_zero & page_mask,
 529                             short_at(should_be_zero));
 530                         report_failure(205);
 531                 }
 532 }
 533
 534
 535 void
 536 domapread(unsigned offset, unsigned size)
 537 {
 538         unsigned pg_offset;
 539         unsigned map_size;
 540         char    *p;
 541
 542         offset -= offset % readbdy;
 543         if (size == 0) {
 544                 if (!quiet && testcalls > simulatedopcount)
 545                         prt("skipping zero size read\n");
 546                 log4(OP_SKIPPED, OP_MAPREAD, offset, size);
 547                 return;
 548         }
 549         if (size + offset > file_size) {
 550                 if (!quiet && testcalls > simulatedopcount)
 551                         prt("skipping seek/read past end of file\n");
 552                 log4(OP_SKIPPED, OP_MAPREAD, offset, size);
 553                 return;
 554         }
 555
 556         log4(OP_MAPREAD, offset, size, 0);
 557
 558         if (testcalls <= simulatedopcount)
 559                 return;
 560
 561         if (!quiet &&
 562                 ((progressinterval && testcalls % progressinterval == 0) ||
 563                        (debug &&
 564                        (monitorstart == -1 ||
 565                         (offset + size > monitorstart &&
 566                         (monitorend == -1 || offset <= monitorend))))))
 567                 prt("%lu mapread\t0x%x thru\t0x%x\t(0x%x bytes)\n", testcalls,
 568                     offset, offset + size - 1, size);
 569
 570         pg_offset = offset & PAGE_MASK;
 571         map_size  = pg_offset + size;
 572
 573         if ((p = (char *)mmap(0, map_size, PROT_READ, MAP_SHARED, fd,
 574                               (off_t)(offset - pg_offset))) == (char *)-1) {
 575                 prterr("domapread: mmap");
 576                 report_failure(190);
 577         }
 578         memcpy(temp_buf, p + pg_offset, size);
 579
 580         check_eofpage("Read", offset, p, size);
 581
 582         if (munmap(p, map_size) != 0) {
 583                 prterr("domapread: munmap");
 584                 report_failure(191);
 585         }
 586
 587         check_buffers(offset, size);
 588 }
 589
 590
 591 void
 592 gendata(char *original_buf, char *good_buf, unsigned offset, unsigned size)
 593 {
 594         while (size--) {
 595                 good_buf[offset] = testcalls % 256;
 596                 if (offset % 2)
 597                         good_buf[offset] += original_buf[offset];
 598                 offset++;
 599         }
 600 }
 601
 602
 603 void
 604 dowrite(unsigned offset, unsigned size)
 605 {
 606         off_t ret;
 607         unsigned iret;
 608
 609         offset -= offset % writebdy;
 610         if (o_direct)
 611                 size -= size % writebdy;
 612         if (size == 0) {
 613                 if (!quiet && testcalls > simulatedopcount && !o_direct)
 614                         prt("skipping zero size write\n");
 615                 log4(OP_SKIPPED, OP_WRITE, offset, size);
 616                 return;
 617         }
 618
 619         log4(OP_WRITE, offset, size, file_size);
 620
 621         gendata(original_buf, good_buf, offset, size);
 622         if (file_size < offset + size) {
 623                 if (file_size < offset)
 624                         memset(good_buf + file_size, '\0', offset - file_size);
 625                 file_size = offset + size;
 626                 if (lite) {
 627                         warn("Lite file size bug in fsx!");
 628                         report_failure(149);
 629                 }
 630         }
 631
 632         if (testcalls <= simulatedopcount)
 633                 return;
 634
 635         if (!quiet &&
 636                 ((progressinterval && testcalls % progressinterval == 0) ||
 637                        (debug &&
 638                        (monitorstart == -1 ||
 639                         (offset + size > monitorstart &&
 640                         (monitorend == -1 || offset <= monitorend))))))
 641                 prt("%lu write\t0x%x thru\t0x%x\t(0x%x bytes)\n", testcalls,
 642                     offset, offset + size - 1, size);
 643         ret = lseek(fd, (off_t)offset, SEEK_SET);
 644         if (ret == (off_t)-1) {
 645                 prterr("dowrite: lseek");
 646                 report_failure(150);
 647         }
 648         iret = fsxwrite(fd, good_buf + offset, size, offset);
 649         if (iret != size) {
 650                 if (iret == -1)
 651                         prterr("dowrite: write");
 652                 else
 653                         prt("short write: 0x%x bytes instead of 0x%x\n",
 654                             iret, size);
 655                 report_failure(151);
 656         }
 657         if (do_fsync) {
 658                 if (fsync(fd)) {
 659                         prt("fsync() failed: %s\n", strerror(errno));
 660                         report_failure(152);
 661                 }
 662         }
 663         if (flush) {
 664                 doflush(offset, size);
 665         }
 666 }
 667
 668
 669 void
 670 domapwrite(unsigned offset, unsigned size)
 671 {
 672         unsigned pg_offset;
 673         unsigned map_size;
 674         off_t    cur_filesize;
 675         char    *p;
 676
 677         offset -= offset % writebdy;
 678         if (size == 0) {
 679                 if (!quiet && testcalls > simulatedopcount)
 680                         prt("skipping zero size write\n");
 681                 log4(OP_SKIPPED, OP_MAPWRITE, offset, size);
 682                 return;
 683         }
 684         cur_filesize = file_size;
 685
 686         log4(OP_MAPWRITE, offset, size, 0);
 687
 688         gendata(original_buf, good_buf, offset, size);
 689         if (file_size < offset + size) {
 690                 if (file_size < offset)
 691                         memset(good_buf + file_size, '\0', offset - file_size);
 692                 file_size = offset + size;
 693                 if (lite) {
 694                         warn("Lite file size bug in fsx!");
 695                         report_failure(200);
 696                 }
 697         }
 698
 699         if (testcalls <= simulatedopcount)
 700                 return;
 701
 702         if (!quiet &&
 703                 ((progressinterval && testcalls % progressinterval == 0) ||
 704                        (debug &&
 705                        (monitorstart == -1 ||
 706                         (offset + size > monitorstart &&
 707                         (monitorend == -1 || offset <= monitorend))))))
 708                 prt("%lu mapwrite\t0x%x thru\t0x%x\t(0x%x bytes)\n", testcalls,
 709                     offset, offset + size - 1, size);
 710
 711         if (file_size > cur_filesize) {
 712                 if (ftruncate(fd, file_size) == -1) {
 713                         prterr("domapwrite: ftruncate");
 714                         exit(201);
 715                 }
 716         }
 717         pg_offset = offset & PAGE_MASK;
 718         map_size  = pg_offset + size;
 719
 720         if ((p = (char *)mmap(0, map_size, PROT_READ | PROT_WRITE,
 721                               MAP_FILE | MAP_SHARED, fd,
 722                               (off_t)(offset - pg_offset))) == (char *)-1) {
 723                 prterr("domapwrite: mmap");
 724                 report_failure(202);
 725         }
 726         memcpy(p + pg_offset, good_buf + offset, size);
 727         if (msync(p, map_size, 0) != 0) {
 728                 prterr("domapwrite: msync");
 729                 report_failure(203);
 730         }
 731
 732         check_eofpage("Write", offset, p, size);
 733
 734         if (munmap(p, map_size) != 0) {
 735                 prterr("domapwrite: munmap");
 736                 report_failure(204);
 737         }
 738 }
 739
 740
 741 void
 742 dotruncate(unsigned size)
 743 {
 744         int oldsize = file_size;
 745
 746         size -= size % truncbdy;
 747         if (size > biggest) {
 748                 biggest = size;
 749                 if (!quiet && testcalls > simulatedopcount)
 750                         prt("truncating to largest ever: 0x%x\n", size);
 751         }
 752
 753         log4(OP_TRUNCATE, size, (unsigned)file_size, 0);
 754
 755         if (size > file_size)
 756                 memset(good_buf + file_size, '\0', size - file_size);
 757         file_size = size;
 758
 759         if (testcalls <= simulatedopcount)
 760                 return;
 761
 762         if ((progressinterval && testcalls % progressinterval == 0) ||
 763             (debug && (monitorstart == -1 || monitorend == -1 ||
 764                       size <= monitorend)))
 765                 prt("%lu trunc\tfrom 0x%x to 0x%x\n", testcalls, oldsize, size);
 766         if (ftruncate(fd, (off_t)size) == -1) {
 767                 prt("ftruncate1: %x\n", size);
 768                 prterr("dotruncate: ftruncate");
 769                 report_failure(160);
 770         }
 771 }
 772
 773
 774 void
 775 writefileimage()
 776 {
 777         ssize_t iret;
 778
 779         if (lseek(fd, (off_t)0, SEEK_SET) == (off_t)-1) {
 780                 prterr("writefileimage: lseek");
 781                 report_failure(171);
 782         }
 783         iret = write(fd, good_buf, file_size);
 784         if ((off_t)iret != file_size) {
 785                 if (iret == -1)
 786                         prterr("writefileimage: write");
 787                 else
 788                         prt("short write: 0x%x bytes instead of 0x%llx\n",
 789                             iret, (unsigned long long)file_size);
 790                 report_failure(172);
 791         }
 792         if (lite ? 0 : ftruncate(fd, file_size) == -1) {
 793                 prt("ftruncate2: %llx\n", (unsigned long long)file_size);
 794                 prterr("writefileimage: ftruncate");
 795                 report_failure(173);
 796         }
 797 }
 798
 799
 800 void
 801 docloseopen(void)
 802 {
 803         if (testcalls <= simulatedopcount)
 804                 return;
 805
 806         if (debug)
 807                 prt("%lu close/open\n", testcalls);
 808         if (close(fd)) {
 809                 prterr("docloseopen: close");
 810                 report_failure(180);
 811         }
 812         fd = open(fname, O_RDWR|o_direct, 0);
 813         if (fd < 0) {
 814                 prterr("docloseopen: open");
 815                 report_failure(181);
 816         }
 817 }
 818
 819
 820 void
 821 test(void)
 822 {
 823         unsigned long   offset;
 824         unsigned long   size = maxoplen;
 825         unsigned long   rv = random();
 826         unsigned long   op = rv % (3 + !lite + mapped_writes);
 827
 828         /* turn off the map read if necessary */
 829
 830         if (op == 2 && !mapped_reads)
 831             op = 0;
 832
 833         if (simulatedopcount > 0 && testcalls == simulatedopcount)
 834                 writefileimage();
 835
 836         testcalls++;
 837
 838         if (closeprob)
 839                 closeopen = (rv >> 3) < (1 << 28) / closeprob;
 840
 841         if (debugstart > 0 && testcalls >= debugstart)
 842                 debug = 1;
 843
 844         if (!quiet && testcalls < simulatedopcount && testcalls % 100000 == 0)
 845                 prt("%lu...\n", testcalls);
 846
 847         /*
 848          * READ:        op = 0
 849          * WRITE:       op = 1
 850          * MAPREAD:     op = 2
 851          * TRUNCATE:    op = 3
 852          * MAPWRITE:    op = 3 or 4
 853          */
 854         if (lite ? 0 : op == 3 && (style & 1) == 0) /* vanilla truncate? */
 855                 dotruncate(random() % maxfilelen);
 856         else {
 857                 if (randomoplen)
 858                         size = random() % (maxoplen+1);
 859                 if (lite ? 0 : op == 3)
 860                         dotruncate(size);
 861                 else {
 862                         offset = random();
 863                         if (op == 1 || op == (lite ? 3 : 4)) {
 864                                 offset %= maxfilelen;
 865                                 if (offset + size > maxfilelen)
 866                                         size = maxfilelen - offset;
 867                                 if (op != 1)
 868                                         domapwrite(offset, size);
 869                                 else
 870                                         dowrite(offset, size);
 871                         } else {
 872                                 if (file_size)
 873                                         offset %= file_size;
 874                                 else
 875                                         offset = 0;
 876                                 if (offset + size > file_size)
 877                                         size = file_size - offset;
 878                                 if (op != 0)
 879                                         domapread(offset, size);
 880                                 else
 881                                         doread(offset, size);
 882                         }
 883                 }
 884         }
 885         if (sizechecks && testcalls > simulatedopcount)
 886                 check_size();
 887         if (closeopen)
 888                 docloseopen();
 889 }
 890
 891
 892 void
 893 cleanup(sig)
 894         int     sig;
 895 {
 896         if (sig)
 897                 prt("signal %d\n", sig);
 898         prt("testcalls = %lu\n", testcalls);
 899         exit(sig);
 900 }
 901
 902
 903 void
 904 usage(void)
 905 {
 906         fprintf(stdout, "usage: %s",
 907                 "fsx [-dnqxALOWZ] [-b opnum] [-c Prob] [-l flen] [-m start:end] [-o oplen] [-p progressinterval] [-r readbdy] [-s style] [-t truncbdy] [-w writebdy] [-D startingop] [-N numops] [-P dirpath] [-S seed] fname\n\
 908         -b opnum: beginning operation number (default 1)\n\
 909         -c P: 1 in P chance of file close+open at each op (default infinity)\n\
 910         -d: debug output for all operations\n\
 911         -f flush and invalidate cache after I/O\n\
 912         -l flen: the upper bound on file size (default 262144)\n\
 913         -m startop:endop: monitor (print debug output) specified byte range (default 0:infinity)\n\
 914         -n: no verifications of file size\n\
 915         -o oplen: the upper bound on operation size (default 65536)\n\
 916         -p progressinterval: debug output at specified operation interval\n\
 917         -q: quieter operation\n\
 918         -r readbdy: 4096 would make reads page aligned (default 1)\n\
 919         -s style: 1 gives smaller truncates (default 0)\n\
 920         -t truncbdy: 4096 would make truncates page aligned (default 1)\n\
 921         -w writebdy: 4096 would make writes page aligned (default 1)\n\
 922         -x: preallocate file space before starting, XFS only (default 0)\n\
 923         -y synchronize changes to a file\n"
 924
 925 #ifdef AIO
 926 "       -A: Use the AIO system calls\n"
 927 #endif
 928 "       -D startingop: debug output starting at specified operation\n\
 929         -L: fsxLite - no file creations & no file size changes\n\
 930         -N numops: total # operations to do (default infinity)\n\
 931         -O: use oplen (see -o flag) for every op (default random)\n\
 932         -P: save .fsxlog and .fsxgood files in dirpath (default ./)\n\
 933         -S seed: for random # generator (default 1) 0 gets timestamp\n\
 934         -W: mapped write operations DISabled\n\
 935         -R: read() system calls only (mapped reads disabled)\n\
 936         -Z: O_DIRECT (use -R, -W, -r and -w too)\n\
 937         fname: this filename is REQUIRED (no default)\n");
 938         exit(90);
 939 }
 940
 941
 942 int
 943 getnum(char *s, char **e)
 944 {
 945         int ret;
 946
 947         *e = (char *) 0;
 948         ret = strtol(s, e, 0);
 949         if (*e)
 950                 switch (**e) {
 951                 case 'b':
 952                 case 'B':
 953                         ret *= 512;
 954                         *e = *e + 1;
 955                         break;
 956                 case 'k':
 957                 case 'K':
 958                         ret *= 1024;
 959                         *e = *e + 1;
 960                         break;
 961                 case 'm':
 962                 case 'M':
 963                         ret *= 1024*1024;
 964                         *e = *e + 1;
 965                         break;
 966                 case 'w':
 967                 case 'W':
 968                         ret *= 4;
 969                         *e = *e + 1;
 970                         break;
 971                 }
 972         return (ret);
 973 }
 974
 975 #ifdef AIO
 976
 977 #define QSZ     1024
 978 io_context_t    io_ctx;
 979 struct iocb     iocb;
 980
 981 int aio_setup()
 982 {
 983         int ret;
 984         ret = io_queue_init(QSZ, &io_ctx);
 985         if (ret != 0) {
 986                 fprintf(stderr, "aio_setup: io_queue_init failed: %s\n",
 987                         strerror(ret));
 988                 return(-1);
 989         }
 990         return(0);
 991 }
 992
 993 int
 994 __aio_rw(int rw, int fd, char *buf, unsigned len, unsigned offset)
 995 {
 996         struct io_event event;
 997         static struct timespec ts;
 998         struct iocb *iocbs[] = { &iocb };
 999         int ret;
1000         long res;
1001
1002         if (rw == READ) {
1003                 io_prep_pread(&iocb, fd, buf, len, offset);
1004         } else {
1005                 io_prep_pwrite(&iocb, fd, buf, len, offset);
1006         }
1007
1008         ts.tv_sec = 30;
1009         ts.tv_nsec = 0;
1010         ret = io_submit(io_ctx, 1, iocbs);
1011         if (ret != 1) {
1012                 fprintf(stderr, "errcode=%d\n", ret);
1013                 fprintf(stderr, "aio_rw: io_submit failed: %s\n",
1014                                 strerror(ret));
1015                 goto out_error;
1016         }
1017
1018         ret = io_getevents(io_ctx, 1, 1, &event, &ts);
1019         if (ret != 1) {
1020                 if (ret == 0)
1021                         fprintf(stderr, "aio_rw: no events available\n");
1022                 else {
1023                         fprintf(stderr, "errcode=%d\n", -ret);
1024                         fprintf(stderr, "aio_rw: io_getevents failed: %s\n",
1025                                         strerror(-ret));
1026                 }
1027                 goto out_error;
1028         }
1029         if (len != event.res) {
1030                 /*
1031                  * The b0rked libaio defines event.res as unsigned.
1032                  * However the kernel strucuture has it signed,
1033                  * and it's used to pass negated error value.
1034                  * Till the library is fixed use the temp var.
1035                  */
1036                 res = (long)event.res;
1037                 if (res >= 0)
1038                         fprintf(stderr, "bad io length: %lu instead of %u\n",
1039                                         res, len);
1040                 else {
1041                         fprintf(stderr, "errcode=%d\n", -res);
1042                         fprintf(stderr, "aio_rw: async io failed: %s\n",
1043                                         strerror(-res));
1044                         ret = res;
1045                         goto out_error;
1046                 }
1047
1048         }
1049         return event.res;
1050
1051 out_error:
1052         /*
1053          * The caller expects error return in traditional libc
1054          * convention, i.e. -1 and the errno set to error.
1055          */
1056         errno = -ret;
1057         return -1;
1058 }
1059
1060 int aio_rw(int rw, int fd, char *buf, unsigned len, unsigned offset)
1061 {
1062         int ret;
1063
1064         if (aio) {
1065                 ret = __aio_rw(rw, fd, buf, len, offset);
1066         } else {
1067                 if (rw == READ)
1068                         ret = read(fd, buf, len);
1069                 else
1070                         ret = write(fd, buf, len);
1071         }
1072         return ret;
1073 }
1074
1075 #endif
1076
1077 int
1078 main(int argc, char **argv)
1079 {
1080         int     i, style, ch;
1081         char    *endp;
1082         char goodfile[1024];
1083         char logfile[1024];
1084
1085         goodfile[0] = 0;
1086         logfile[0] = 0;
1087
1088         page_size = getpagesize();
1089         page_mask = page_size - 1;
1090         mmap_mask = page_mask;
1091
1092
1093         setvbuf(stdout, (char *)0, _IOLBF, 0); /* line buffered stdout */
1094
1095         while ((ch = getopt(argc, argv, "b:c:dfl:m:no:p:qr:s:t:w:xyAD:LN:OP:RS:WZ"))
1096                != EOF)
1097                 switch (ch) {
1098                 case 'b':
1099                         simulatedopcount = getnum(optarg, &endp);
1100                         if (!quiet)
1101                                 fprintf(stdout, "Will begin at operation %ld\n",
1102                                         simulatedopcount);
1103                         if (simulatedopcount == 0)
1104                                 usage();
1105                         simulatedopcount -= 1;
1106                         break;
1107                 case 'c':
1108                         closeprob = getnum(optarg, &endp);
1109                         if (!quiet)
1110                                 fprintf(stdout,
1111                                         "Chance of close/open is 1 in %d\n",
1112                                         closeprob);
1113                         if (closeprob <= 0)
1114                                 usage();
1115                         break;
1116                 case 'd':
1117                         debug = 1;
1118                         break;
1119                 case 'f':
1120                         flush = 1;
1121                         break;
1122                 case 'l':
1123                         maxfilelen = getnum(optarg, &endp);
1124                         if (maxfilelen <= 0)
1125                                 usage();
1126                         break;
1127                 case 'm':
1128                         monitorstart = getnum(optarg, &endp);
1129                         if (monitorstart < 0)
1130                                 usage();
1131                         if (!endp || *endp++ != ':')
1132                                 usage();
1133                         monitorend = getnum(endp, &endp);
1134                         if (monitorend < 0)
1135                                 usage();
1136                         if (monitorend == 0)
1137                                 monitorend = -1; /* aka infinity */
1138                         debug = 1;
1139                 case 'n':
1140                         sizechecks = 0;
1141                         break;
1142                 case 'o':
1143                         maxoplen = getnum(optarg, &endp);
1144                         if (maxoplen <= 0)
1145                                 usage();
1146                         break;
1147                 case 'p':
1148                         progressinterval = getnum(optarg, &endp);
1149                         if (progressinterval == 0)
1150                                 usage();
1151                         break;
1152                 case 'q':
1153                         quiet = 1;
1154                         break;
1155                 case 'r':
1156                         readbdy = getnum(optarg, &endp);
1157                         if (readbdy <= 0)
1158                                 usage();
1159                         break;
1160                 case 's':
1161                         style = getnum(optarg, &endp);
1162                         if (style < 0 || style > 1)
1163                                 usage();
1164                         break;
1165                 case 't':
1166                         truncbdy = getnum(optarg, &endp);
1167                         if (truncbdy <= 0)
1168                                 usage();
1169                         break;
1170                 case 'w':
1171                         writebdy = getnum(optarg, &endp);
1172                         if (writebdy <= 0)
1173                                 usage();
1174                         break;
1175                 case 'x':
1176                         prealloc = 1;
1177                         break;
1178                 case 'y':
1179                         do_fsync = 1;
1180                         break;
1181                 case 'A':
1182                         aio = 1;
1183                         break;
1184                 case 'D':
1185                         debugstart = getnum(optarg, &endp);
1186                         if (debugstart < 1)
1187                                 usage();
1188                         break;
1189                 case 'L':
1190                         lite = 1;
1191                         break;
1192                 case 'N':
1193                         numops = getnum(optarg, &endp);
1194                         if (numops < 0)
1195                                 usage();
1196                         break;
1197                 case 'O':
1198                         randomoplen = 0;
1199                         break;
1200                 case 'P':
1201                         strncpy(goodfile, optarg, sizeof(goodfile));
1202                         strcat(goodfile, "/");
1203                         strncpy(logfile, optarg, sizeof(logfile));
1204                         strcat(logfile, "/");
1205                         break;
1206                 case 'R':
1207                         mapped_reads = 0;
1208                         break;
1209                 case 'S':
1210                         seed = getnum(optarg, &endp);
1211                         if (seed == 0)
1212                                 seed = time(0) % 10000;
1213                         if (!quiet)
1214                                 fprintf(stdout, "Seed set to %d\n", seed);
1215                         if (seed < 0)
1216                                 usage();
1217                         break;
1218                 case 'W':
1219                         mapped_writes = 0;
1220                         if (!quiet)
1221                                 fprintf(stdout, "mapped writes DISABLED\n");
1222                         break;
1223                 case 'Z':
1224                         o_direct = O_DIRECT;
1225                         break;
1226                 default:
1227                         usage();
1228                         /* NOTREACHED */
1229                 }
1230         argc -= optind;
1231         argv += optind;
1232         if (argc != 1)
1233                 usage();
1234         fname = argv[0];
1235
1236         signal(SIGHUP,  cleanup);
1237         signal(SIGINT,  cleanup);
1238         signal(SIGPIPE, cleanup);
1239         signal(SIGALRM, cleanup);
1240         signal(SIGTERM, cleanup);
1241         signal(SIGXCPU, cleanup);
1242         signal(SIGXFSZ, cleanup);
1243         signal(SIGVTALRM,       cleanup);
1244         signal(SIGUSR1, cleanup);
1245         signal(SIGUSR2, cleanup);
1246
1247         initstate(seed, state, 256);
1248         setstate(state);
1249         fd = open(fname,
1250                 O_RDWR|(lite ? 0 : O_CREAT|O_TRUNC)|o_direct, 0666);
1251         if (fd < 0) {
1252                 prterr(fname);
1253                 exit(91);
1254         }
1255 #ifdef XFS
1256         if (prealloc) {
1257                 xfs_flock64_t   resv = { 0 };
1258 #ifdef HAVE_XFS_PLATFORM_DEFS_H
1259                 if (!platform_test_xfs_fd(fd)) {
1260                         prterr(fname);
1261                         fprintf(stderr, "main: cannot prealloc, non XFS\n");
1262                         exit(96);
1263                 }
1264 #endif
1265                 resv.l_len = maxfilelen;
1266                 if ((xfsctl(fname, fd, XFS_IOC_RESVSP, &resv)) < 0) {
1267                         prterr(fname);
1268                         exit(97);
1269                 }
1270         }
1271 #endif
1272         strncat(goodfile, fname, 256);
1273         strcat (goodfile, ".fsxgood");
1274         fsxgoodfd = open(goodfile, O_RDWR|O_CREAT|O_TRUNC, 0666);
1275         if (fsxgoodfd < 0) {
1276                 prterr(goodfile);
1277                 exit(92);
1278         }
1279         strncat(logfile, fname, 256);
1280         strcat (logfile, ".fsxlog");
1281         fsxlogf = fopen(logfile, "w");
1282         if (fsxlogf == NULL) {
1283                 prterr(logfile);
1284                 exit(93);
1285         }
1286
1287 #ifdef AIO
1288         if (aio)
1289                 aio_setup();
1290 #endif
1291
1292         if (lite) {
1293                 off_t ret;
1294                 file_size = maxfilelen = lseek(fd, (off_t)0, SEEK_END);
1295                 if (file_size == (off_t)-1) {
1296                         prterr(fname);
1297                         warn("main: lseek eof");
1298                         exit(94);
1299                 }
1300                 ret = lseek(fd, (off_t)0, SEEK_SET);
1301                 if (ret == (off_t)-1) {
1302                         prterr(fname);
1303                         warn("main: lseek 0");
1304                         exit(95);
1305                 }
1306         }
1307         original_buf = (char *) malloc(maxfilelen);
1308         for (i = 0; i < maxfilelen; i++)
1309                 original_buf[i] = random() % 256;
1310         good_buf = (char *) malloc(maxfilelen + writebdy);
1311         good_buf = round_up(good_buf, writebdy, 0);
1312         memset(good_buf, '\0', maxfilelen);
1313         temp_buf = (char *) malloc(maxoplen + readbdy);
1314         temp_buf = round_up(temp_buf, readbdy, 0);
1315         memset(temp_buf, '\0', maxoplen);
1316         if (lite) {     /* zero entire existing file */
1317                 ssize_t written;
1318
1319                 written = write(fd, good_buf, (size_t)maxfilelen);
1320                 if (written != maxfilelen) {
1321                         if (written == -1) {
1322                                 prterr(fname);
1323                                 warn("main: error on write");
1324                         } else
1325                                 warn("main: short write, 0x%x bytes instead "
1326                                         "of 0x%lx\n",
1327                                         (unsigned)written,
1328                                         maxfilelen);
1329                         exit(98);
1330                 }
1331         } else
1332                 check_trunc_hack();
1333
1334         while (numops == -1 || numops--)
1335                 test();
1336
1337         if (close(fd)) {
1338                 prterr("close");
1339                 report_failure(99);
1340         }
1341         prt("All operations completed A-OK!\n");
1342
1343         exit(0);
1344         return 0;
1345 }