coregrind/m_libcproc.c

   1
   2 /*--------------------------------------------------------------------*/
   3 /*--- Process-related libc stuff.                     m_libcproc.c ---*/
   4 /*--------------------------------------------------------------------*/
   5
   6 /*
   7    This file is part of Valgrind, a dynamic binary instrumentation
   8    framework.
   9
  10    Copyright (C) 2000-2017 Julian Seward
  11       jseward@acm.org
  12
  13    This program is free software; you can redistribute it and/or
  14    modify it under the terms of the GNU General Public License as
  15    published by the Free Software Foundation; either version 2 of the
  16    License, or (at your option) any later version.
  17
  18    This program is distributed in the hope that it will be useful, but
  19    WITHOUT ANY WARRANTY; without even the implied warranty of
  20    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  21    General Public License for more details.
  22
  23    You should have received a copy of the GNU General Public License
  24    along with this program; if not, see <http://www.gnu.org/licenses/>.
  25
  26    The GNU General Public License is contained in the file COPYING.
  27 */
  28
  29 #include "pub_core_basics.h"
  30 #include "pub_core_machine.h"    // For VG_(machine_get_VexArchInfo)
  31 #include "pub_core_vki.h"
  32 #include "pub_core_vkiscnums.h"
  33 #include "pub_core_libcbase.h"
  34 #include "pub_core_libcassert.h"
  35 #include "pub_core_libcfile.h"
  36 #include "pub_core_libcprint.h"
  37 #include "pub_core_libcproc.h"
  38 #include "pub_core_libcsignal.h"
  39 #include "pub_core_seqmatch.h"
  40 #include "pub_core_mallocfree.h"
  41 #include "pub_core_signals.h"
  42 #include "pub_core_syscall.h"
  43 #include "pub_core_xarray.h"
  44 #include "pub_core_clientstate.h"
  45 #include "pub_core_debuglog.h"   // VG_(debugLog)
  46
  47 #if defined(VGO_darwin)
  48 /* --- !!! --- EXTERNAL HEADERS start --- !!! --- */
  49 #include <mach/mach.h>   /* mach_thread_self */
  50 /* --- !!! --- EXTERNAL HEADERS end --- !!! --- */
  51 #endif
  52
  53 /* IMPORTANT: on Darwin it is essential to use the _nocancel versions
  54    of syscalls rather than the vanilla version, if a _nocancel version
  55    is available.  See docs/internals/Darwin-notes.txt for the reason
  56    why. */
  57
  58 /* ---------------------------------------------------------------------
  59    Command line and environment stuff
  60    ------------------------------------------------------------------ */
  61
  62 /* As deduced from sp_at_startup, the client's argc, argv[] and
  63    envp[] as extracted from the client's stack at startup-time. */
  64 HChar** VG_(client_envp) = NULL;
  65
  66 /* Path to library directory */
  67 const HChar *VG_(libdir) = VG_LIBDIR;
  68
  69 const HChar *VG_(LD_PRELOAD_var_name) =
  70 #if defined(VGO_linux) || defined(VGO_solaris) || defined(VGO_freebsd)
  71    "LD_PRELOAD";
  72 #elif defined(VGO_darwin)
  73    "DYLD_INSERT_LIBRARIES";
  74 #else
  75 #  error Unknown OS
  76 #endif
  77
  78 /* We do getenv without libc's help by snooping around in
  79    VG_(client_envp) as determined at startup time. */
  80 HChar *VG_(getenv)(const HChar *varname)
  81 {
  82    Int i, n;
  83    vg_assert( VG_(client_envp) );
  84    n = VG_(strlen)(varname);
  85    for (i = 0; VG_(client_envp)[i] != NULL; i++) {
  86       HChar* s = VG_(client_envp)[i];
  87       if (VG_(strncmp)(varname, s, n) == 0 && s[n] == '=') {
  88          return & s[n+1];
  89       }
  90    }
  91    return NULL;
  92 }
  93
  94 /* If free_fn is not NULL, it is called on "unset" environment variable. */
  95 void  VG_(env_unsetenv) ( HChar **env, const HChar *varname,
  96                           void (*free_fn) (void *) )
  97 {
  98    HChar **from, **to;
  99    vg_assert(env);
 100    vg_assert(varname);
 101    to = NULL;
 102    Int len = VG_(strlen)(varname);
 103
 104    for (from = to = env; from && *from; from++) {
 105       if (!(VG_(strncmp)(varname, *from, len) == 0 && (*from)[len] == '=')) {
 106          *to = *from;
 107          to++;
 108       } else if (free_fn != NULL) {
 109          free_fn(*from);
 110       }
 111    }
 112    *to = *from;
 113 }
 114
 115 /* set the environment; returns the old env if a new one was allocated */
 116 HChar **VG_(env_setenv) ( HChar ***envp, const HChar* varname,
 117                           const HChar *val )
 118 {
 119    HChar **env = (*envp);
 120    HChar **cpp;
 121    Int len = VG_(strlen)(varname);
 122    HChar *valstr = VG_(malloc)("libcproc.es.1", len + VG_(strlen)(val) + 2);
 123    HChar **oldenv = NULL;
 124
 125    VG_(sprintf)(valstr, "%s=%s", varname, val);
 126
 127    for (cpp = env; cpp && *cpp; cpp++) {
 128       if (VG_(strncmp)(varname, *cpp, len) == 0 && (*cpp)[len] == '=') {
 129          *cpp = valstr;
 130          return oldenv;
 131       }
 132    }
 133
 134    if (env == NULL) {
 135       env = VG_(malloc)("libcproc.es.2", sizeof(HChar *) * 2);
 136       env[0] = valstr;
 137       env[1] = NULL;
 138
 139       *envp = env;
 140
 141    }  else {
 142       Int envlen = (cpp-env) + 2;
 143       HChar **newenv = VG_(malloc)("libcproc.es.3", envlen * sizeof(HChar *));
 144
 145       for (cpp = newenv; *env; )
 146          *cpp++ = *env++;
 147       *cpp++ = valstr;
 148       *cpp++ = NULL;
 149
 150       oldenv = *envp;
 151
 152       *envp = newenv;
 153    }
 154
 155    return oldenv;
 156 }
 157
 158
 159 /* Walk through a colon-separated environment variable, and remove the
 160    entries which match remove_pattern.  It slides everything down over
 161    the removed entries, and pads the remaining space with '\0'.  It
 162    modifies the entries in place (in the client address space), but it
 163    shouldn't matter too much, since we only do this just before an
 164    execve().
 165
 166    This is also careful to mop up any excess ':'s, since empty strings
 167    delimited by ':' are considered to be '.' in a path.
 168 */
 169 static void mash_colon_env(HChar *varp, const HChar *remove_pattern)
 170 {
 171    HChar *const start = varp;
 172    HChar *entry_start = varp;
 173    HChar *output = varp;
 174
 175    if (varp == NULL)
 176       return;
 177
 178    while(*varp) {
 179       if (*varp == ':') {
 180          HChar prev;
 181          Bool match;
 182
 183          /* This is a bit subtle: we want to match against the entry
 184             we just copied, because it may have overlapped with
 185             itself, junking the original. */
 186
 187          prev = *output;
 188          *output = '\0';
 189
 190          match = VG_(string_match)(remove_pattern, entry_start);
 191
 192          *output = prev;
 193
 194          if (match) {
 195             output = entry_start;
 196             varp++;                     /* skip ':' after removed entry */
 197          } else
 198             entry_start = output+1;     /* entry starts after ':' */
 199       }
 200
 201       if (*varp)
 202          *output++ = *varp++;
 203    }
 204
 205    /* make sure last entry is nul terminated */
 206    *output = '\0';
 207
 208    /* match against the last entry */
 209    if (VG_(string_match)(remove_pattern, entry_start)) {
 210       output = entry_start;
 211       if (output > start) {
 212          /* remove trailing ':' */
 213          output--;
 214          vg_assert(*output == ':');
 215       }
 216    }
 217
 218    /* pad out the left-overs with '\0' */
 219    while(output < varp)
 220       *output++ = '\0';
 221 }
 222
 223
 224 /* Removes all the Valgrind-added stuff from the passed environment.  Used
 225    when starting child processes, so they don't see that added stuff.
 226    If the ro_strings option is set to True then all strings referenced by envp
 227    are considered read-only, which means they will be duplicated before they
 228    are modified.
 229    If free_fn is not NULL, it is called on "unset" environment variables. */
 230 void VG_(env_remove_valgrind_env_stuff)(HChar** envp, Bool ro_strings,
 231                                         void (*free_fn) (void *) )
 232 {
 233    Int i;
 234    HChar* ld_preload_str = NULL;
 235    HChar* ld_library_path_str = NULL;
 236    HChar* dyld_insert_libraries_str = NULL;
 237    HChar* buf;
 238
 239    // Find LD_* variables
 240    // DDD: should probably conditionally compiled some of this:
 241    // - LD_LIBRARY_PATH is universal?
 242    // - LD_PRELOAD is on Linux, not on Darwin, not sure about AIX
 243    // - DYLD_INSERT_LIBRARIES and DYLD_SHARED_REGION are Darwin-only
 244    for (i = 0; envp[i] != NULL; i++) {
 245       if (VG_(strncmp)(envp[i], "LD_PRELOAD=", 11) == 0) {
 246          if (ro_strings)
 247             envp[i] = VG_(strdup)("libcproc.erves.1", envp[i]);
 248          ld_preload_str = &envp[i][11];
 249       }
 250       if (VG_(strncmp)(envp[i], "LD_LIBRARY_PATH=", 16) == 0) {
 251          if (ro_strings)
 252             envp[i] = VG_(strdup)("libcproc.erves.2", envp[i]);
 253          ld_library_path_str = &envp[i][16];
 254       }
 255       if (VG_(strncmp)(envp[i], "DYLD_INSERT_LIBRARIES=", 22) == 0) {
 256          if (ro_strings)
 257             envp[i] = VG_(strdup)("libcproc.erves.3", envp[i]);
 258          dyld_insert_libraries_str = &envp[i][22];
 259       }
 260    }
 261
 262    buf = VG_(malloc)("libcproc.erves.4", VG_(strlen)(VG_(libdir)) + 20);
 263
 264    // Remove Valgrind-specific entries from LD_*.
 265    VG_(sprintf)(buf, "%s*/vgpreload_*.so", VG_(libdir));
 266    mash_colon_env(ld_preload_str, buf);
 267    mash_colon_env(dyld_insert_libraries_str, buf);
 268    VG_(sprintf)(buf, "%s*", VG_(libdir));
 269    mash_colon_env(ld_library_path_str, buf);
 270
 271    // Remove VALGRIND_LAUNCHER variable.
 272    VG_(env_unsetenv)(envp, VALGRIND_LAUNCHER, free_fn);
 273
 274    // Remove DYLD_SHARED_REGION variable.
 275    VG_(env_unsetenv)(envp, "DYLD_SHARED_REGION", free_fn);
 276
 277    // XXX if variable becomes empty, remove it completely?
 278
 279    VG_(free)(buf);
 280 }
 281
 282 /* Resolves filename of VG_(cl_exec_fd) and copies it to the buffer.
 283    Buffer must not be NULL and buf_size must be at least 1.
 284    If buffer is not large enough it is terminated with '\0' only
 285    when 'terminate_with_NUL == True'. */
 286 void VG_(client_fname)(HChar *buffer, SizeT buf_size, Bool terminate_with_NUL)
 287 {
 288    vg_assert(buffer != NULL);
 289    vg_assert(buf_size >= 1);
 290
 291    const HChar *name;
 292    if (VG_(resolve_filename)(VG_(cl_exec_fd), &name)) {
 293       const HChar *n = name + VG_(strlen)(name) - 1;
 294
 295       while (n > name && *n != '/')
 296          n--;
 297       if (n != name)
 298          n++;
 299
 300       VG_(strncpy)(buffer, n, buf_size);
 301       if (terminate_with_NUL)
 302          buffer[buf_size - 1] = '\0';
 303    } else {
 304       buffer[0] = '\0';
 305    }
 306 }
 307
 308 static Bool add_string(HChar *buffer, SizeT *buf_size, const HChar *string)
 309 {
 310    SizeT len = VG_(strlen)(string);
 311    VG_(strncat)(buffer, string, *buf_size);
 312    if (len >= *buf_size - 1) {
 313       *buf_size = 0;
 314       return False;
 315    } else {
 316       *buf_size -= len;
 317       return True;
 318    }
 319 }
 320
 321 /* Concatenates client exename and command line arguments into
 322    the buffer. Buffer must not be NULL and buf_size must be
 323    at least 1. Buffer is always terminated with '\0'. */
 324 void VG_(client_cmd_and_args)(HChar *buffer, SizeT buf_size)
 325 {
 326    vg_assert(buffer != NULL);
 327    vg_assert(buf_size >= 1);
 328
 329    buffer[0] = '\0';
 330
 331    if (add_string(buffer, &buf_size, VG_(args_the_exename)) == False)
 332       return;
 333
 334    Int i;
 335    for (i = 0; i < VG_(sizeXA)(VG_(args_for_client)); i++) {
 336       if (add_string(buffer, &buf_size, " ") == False)
 337          return;
 338
 339       HChar *arg = *(HChar **) VG_(indexXA)(VG_(args_for_client), i);
 340       if (add_string(buffer, &buf_size, arg) == False)
 341          return;
 342    }
 343 }
 344
 345 /* ---------------------------------------------------------------------
 346    Various important syscall wrappers
 347    ------------------------------------------------------------------ */
 348
 349 Int VG_(waitpid)(Int pid, Int *status, Int options)
 350 {
 351 #  if defined(VGO_linux) || defined(VGO_freebsd)
 352    SysRes res = VG_(do_syscall4)(__NR_wait4,
 353                                  pid, (UWord)status, options, 0);
 354    return sr_isError(res) ? -1 : sr_Res(res);
 355 #  elif defined(VGO_darwin)
 356    SysRes res = VG_(do_syscall4)(__NR_wait4_nocancel,
 357                                  pid, (UWord)status, options, 0);
 358    return sr_isError(res) ? -1 : sr_Res(res);
 359 #  elif defined(VGO_solaris)
 360    SysRes res;
 361    vki_idtype_t idtype;
 362    vki_id_t id;
 363    vki_siginfo_t info;
 364
 365    /* We need to do a lot of work here. */
 366
 367    if (pid > 0) {
 368       idtype = VKI_P_PID;
 369       id = pid;
 370    }
 371    else if (pid < -1) {
 372       idtype = VKI_P_PGID;
 373       id = -pid;
 374    }
 375    else if (pid == -1) {
 376       idtype = VKI_P_ALL;
 377       id = 0;
 378    }
 379    else {
 380       idtype = VKI_P_PGID;
 381       res = VG_(do_syscall0)(__NR_getpid);
 382       id = sr_ResHI(res);
 383    }
 384
 385    options |= VKI_WEXITED | VKI_WTRAPPED;
 386
 387    res = VG_(do_syscall4)(__NR_waitsys, idtype, id, (UWord)&info, options);
 388    if (sr_isError(res))
 389       return -1;
 390
 391    if (status) {
 392       Int s = info.si_status & 0xff;
 393
 394       switch (info.si_code) {
 395          case VKI_CLD_EXITED:
 396             s <<= 8;
 397             break;
 398          case VKI_CLD_DUMPED:
 399             s |= VKI_WCOREFLG;
 400             break;
 401          case VKI_CLD_KILLED:
 402             break;
 403          case VKI_CLD_TRAPPED:
 404          case VKI_CLD_STOPPED:
 405             s <<= 8;
 406             s |= VKI_WSTOPFLG;
 407             break;
 408          case VKI_CLD_CONTINUED:
 409             s = VKI_WCONTFLG;
 410             break;
 411       }
 412       *status = s;
 413    }
 414
 415    return info.si_pid;
 416 #  else
 417 #    error Unknown OS
 418 #  endif
 419 }
 420
 421 /* clone the environment */
 422 HChar **VG_(env_clone) ( HChar **oldenv )
 423 {
 424    HChar **oldenvp;
 425    HChar **newenvp;
 426    HChar **newenv;
 427    Int  envlen;
 428
 429    vg_assert(oldenv);
 430    for (oldenvp = oldenv; oldenvp && *oldenvp; oldenvp++);
 431
 432    envlen = oldenvp - oldenv + 1;
 433
 434    newenv = VG_(malloc)("libcproc.ec.1", envlen * sizeof(HChar *));
 435
 436    oldenvp = oldenv;
 437    newenvp = newenv;
 438
 439    while (oldenvp && *oldenvp) {
 440       *newenvp++ = *oldenvp++;
 441    }
 442
 443    *newenvp = *oldenvp;
 444
 445    return newenv;
 446 }
 447
 448 void VG_(execv) ( const HChar* filename, const HChar** argv )
 449 {
 450    HChar** envp;
 451    SysRes res;
 452
 453    envp = VG_(env_clone)(VG_(client_envp));
 454    VG_(env_remove_valgrind_env_stuff)( envp, True /*ro_strings*/, NULL );
 455
 456    res = VG_(do_syscall3)(__NR_execve,
 457                           (UWord)filename, (UWord)argv, (UWord)envp);
 458
 459    VG_(printf)("EXEC failed, errno = %lld\n", (Long)sr_Err(res));
 460 }
 461
 462 /* Spawns a new child. Uses either spawn syscall or fork+execv combo. */
 463 Int VG_(spawn) ( const HChar *filename, const HChar **argv )
 464 {
 465    vg_assert(filename != NULL);
 466    vg_assert(argv != NULL);
 467
 468 #  if defined(VGO_solaris) && defined(SOLARIS_SPAWN_SYSCALL)
 469    HChar **envp = VG_(env_clone)(VG_(client_envp));
 470    for (HChar **p = envp; *p != NULL; p++) {
 471       *p = VG_(strdup)("libcproc.s.1", *p);
 472    }
 473    VG_(env_remove_valgrind_env_stuff)(envp, /* ro_strings */ False, VG_(free));
 474
 475    /* Now combine argv and argp into argenv. */
 476    SizeT argenv_size = 1 + 1;
 477    for (const HChar **p = argv; *p != NULL; p++) {
 478       argenv_size += VG_(strlen)(*p) + 2;
 479    }
 480    for (HChar **p = envp; *p != NULL; p++) {
 481       argenv_size += VG_(strlen)(*p) + 2;
 482    }
 483
 484    HChar *argenv = VG_(malloc)("libcproc.s.2", argenv_size);
 485    HChar *current = argenv;
 486 #  define COPY_CHAR_TO_ARGENV(dst, character)  \
 487       do {                                     \
 488          *(dst) = character;                   \
 489          (dst) += 1;                           \
 490       } while (0)
 491 #  define COPY_STRING_TO_ARGENV(dst, src)        \
 492       do {                                       \
 493          COPY_CHAR_TO_ARGENV(dst, '\1');         \
 494          SizeT src_len = VG_(strlen)((src)) + 1; \
 495          VG_(memcpy)((dst), (src), src_len);     \
 496          (dst) += src_len;                       \
 497       } while (0)
 498
 499    for (const HChar **p = argv; *p != NULL; p++) {
 500       COPY_STRING_TO_ARGENV(current, *p);
 501    }
 502    COPY_CHAR_TO_ARGENV(current, '\0');
 503    for (HChar **p = envp; *p != NULL; p++) {
 504       COPY_STRING_TO_ARGENV(current, *p);
 505    }
 506    COPY_CHAR_TO_ARGENV(current, '\0');
 507    vg_assert(current == argenv + argenv_size);
 508 #  undef COPY_CHAR_TO_ARGENV
 509 #  undef COPY_STRING_TOARGENV
 510
 511    SysRes res = VG_(do_syscall5)(__NR_spawn, (UWord) filename, (UWord) NULL, 0,
 512                                  (UWord) argenv, argenv_size);
 513
 514    VG_(free)(argenv);
 515    for (HChar **p = envp; *p != NULL; p++) {
 516       VG_(free)(*p);
 517    }
 518    VG_(free)(envp);
 519
 520    if (sr_isError(res))
 521       return -1;
 522    return sr_Res(res);
 523
 524 #  else
 525
 526    Int pid = VG_(fork)();
 527    if (pid < 0)
 528       return -1;
 529    if (pid == 0) {
 530       /* child */
 531       VG_(execv)(argv[0], argv);
 532
 533       /* If we're still alive here, execv failed. */
 534       VG_(exit)(1);
 535    } else {
 536       return pid;
 537    }
 538 #  endif /* VGO_solaris && SOLARIS_SPAWN_SYSCALL */
 539 }
 540
 541 /* Return -1 if error, else 0.  NOTE does not indicate return code of
 542    child! */
 543 Int VG_(system) ( const HChar* cmd )
 544 {
 545    Int pid;
 546    if (cmd == NULL)
 547       return 1;
 548
 549    const HChar *argv[4] = { "/bin/sh", "-c", cmd, 0 };
 550    pid = VG_(spawn)(argv[0], argv);
 551    if (pid < 0)
 552       return -1;
 553
 554    vg_assert(pid > 0);
 555    /* parent */
 556    /* We have to set SIGCHLD to its default behaviour in order that
 557       VG_(waitpid) works (at least on AIX).  According to the Linux
 558       man page for waitpid:
 559
 560       POSIX.1-2001 specifies that if the disposition of SIGCHLD is
 561       set to SIG_IGN or the SA_NOCLDWAIT flag is set for SIGCHLD
 562       (see sigaction(2)), then children that terminate do not
 563       become zombies and a call to wait() or waitpid() will block
 564       until all children have terminated, and then fail with errno
 565       set to ECHILD.  (The original POSIX standard left the
 566       behaviour of setting SIGCHLD to SIG_IGN unspecified.)
 567    */
 568    Int ir, zzz;
 569    vki_sigaction_toK_t sa, sa2;
 570    vki_sigaction_fromK_t saved_sa;
 571    VG_(memset)( &sa, 0, sizeof(sa) );
 572    VG_(sigemptyset)(&sa.sa_mask);
 573    sa.ksa_handler = VKI_SIG_DFL;
 574    sa.sa_flags    = 0;
 575    ir = VG_(sigaction)(VKI_SIGCHLD, &sa, &saved_sa);
 576    vg_assert(ir == 0);
 577
 578    zzz = VG_(waitpid)(pid, NULL, 0);
 579
 580    VG_(convert_sigaction_fromK_to_toK)( &saved_sa, &sa2 );
 581    ir = VG_(sigaction)(VKI_SIGCHLD, &sa2, NULL);
 582    vg_assert(ir == 0);
 583    return zzz == -1 ? -1 : 0;
 584 }
 585
 586 Int VG_(sysctl)(Int *name, UInt namelen, void *oldp, SizeT *oldlenp, const void *newp, SizeT newlen)
 587 {
 588    SysRes res;
 589 #  if defined(VGO_darwin) || defined(VGO_freebsd)
 590    res = VG_(do_syscall6)(__NR___sysctl,
 591                            (UWord)name, namelen, (UWord)oldp, (UWord)oldlenp, (UWord)newp, newlen);
 592 #  else
 593    res = VG_(mk_SysRes_Error)(VKI_ENOSYS);
 594 #  endif
 595    return sr_isError(res) ? -1 : sr_Res(res);
 596 }
 597
 598 /* ---------------------------------------------------------------------
 599    Resource limits
 600    ------------------------------------------------------------------ */
 601
 602 /* Support for getrlimit. */
 603 Int VG_(getrlimit) (Int resource, struct vki_rlimit *rlim)
 604 {
 605    SysRes res;
 606    /* res = getrlimit( resource, rlim ); */
 607
 608 #  if defined(__NR_prlimit64) && defined(VKI_RLIM_INFINITY) && defined(VKI_RLIM64_INFINITY)
 609    struct vki_rlimit64 new_rlimit;
 610    res = VG_(do_syscall4)(__NR_prlimit64, 0, resource, 0, (UWord)&new_rlimit);
 611    if (!sr_isError(res)) {
 612       if (new_rlimit.rlim_cur == VKI_RLIM_INFINITY)
 613          new_rlimit.rlim_cur = VKI_RLIM64_INFINITY;
 614       if (new_rlimit.rlim_max == VKI_RLIM_INFINITY)
 615          new_rlimit.rlim_max = VKI_RLIM64_INFINITY;
 616       rlim->rlim_cur = new_rlimit.rlim_cur;
 617       rlim->rlim_max = new_rlimit.rlim_max;
 618       return sr_Res(res);
 619    }
 620    if (sr_Err(res) != VKI_ENOSYS) return -1;
 621 #  endif
 622
 623 #  ifdef __NR_ugetrlimit
 624    res = VG_(do_syscall2)(__NR_ugetrlimit, resource, (UWord)rlim);
 625    if (!sr_isError(res)) return sr_Res(res);
 626    if (sr_Err(res) != VKI_ENOSYS) return -1;
 627 #  endif
 628
 629 #  ifdef __NR_getrlimit
 630    res = VG_(do_syscall2)(__NR_getrlimit, resource, (UWord)rlim);
 631    if (!sr_isError(res)) return sr_Res(res);
 632 #  endif
 633
 634    return -1;
 635 }
 636
 637 /* Support for setrlimit. */
 638 Int VG_(setrlimit) (Int resource, const struct vki_rlimit *rlim)
 639 {
 640    SysRes res;
 641    /* res = setrlimit( resource, rlim ); */
 642
 643 #  ifdef __NR_prlimit64
 644    struct vki_rlimit64 new_rlimit;
 645    new_rlimit.rlim_cur = rlim->rlim_cur;
 646    new_rlimit.rlim_max = rlim->rlim_max;
 647    res = VG_(do_syscall4)(__NR_prlimit64, 0, resource, (UWord)&new_rlimit, 0);
 648    if (!sr_isError(res)) return sr_Res(res);
 649    if (sr_Err(res) != VKI_ENOSYS) return -1;
 650 #  endif
 651
 652 #  ifdef __NR_setrlimit
 653    res = VG_(do_syscall2)(__NR_setrlimit, resource, (UWord)rlim);
 654    if (!sr_isError(res)) return sr_Res(res);
 655    if (sr_Err(res) != VKI_ENOSYS) return -1;
 656 #  endif
 657
 658    return -1;
 659 }
 660
 661 /* Support for prctl. */
 662 Int VG_(prctl) (Int option,
 663                 ULong arg2, ULong arg3, ULong arg4, ULong arg5)
 664 {
 665    SysRes res = VG_(mk_SysRes_Error)(VKI_ENOSYS);
 666 #  if defined(VGO_linux)
 667    /* res = prctl( option, arg2, arg3, arg4, arg5 ); */
 668    res = VG_(do_syscall5)(__NR_prctl, (UWord) option,
 669                           (UWord) arg2, (UWord) arg3, (UWord) arg4,
 670                           (UWord) arg5);
 671 #  endif
 672
 673    return sr_isError(res) ? -1 : sr_Res(res);
 674 }
 675
 676 /* ---------------------------------------------------------------------
 677    pids, etc
 678    ------------------------------------------------------------------ */
 679
 680 Int VG_(gettid)(void)
 681 {
 682 #  if defined(VGO_linux)
 683    SysRes res = VG_(do_syscall0)(__NR_gettid);
 684
 685    if (sr_isError(res) && sr_Res(res) == VKI_ENOSYS) {
 686       HChar pid[16];
 687       /*
 688        * The gettid system call does not exist. The obvious assumption
 689        * to make at this point would be that we are running on an older
 690        * system where the getpid system call actually returns the ID of
 691        * the current thread.
 692        *
 693        * Unfortunately it seems that there are some systems with a kernel
 694        * where getpid has been changed to return the ID of the thread group
 695        * leader but where the gettid system call has not yet been added.
 696        *
 697        * So instead of calling getpid here we use readlink to see where
 698        * the /proc/self link is pointing...
 699        */
 700
 701 #     if defined(VGP_arm64_linux) || defined(VGP_nanomips_linux)
 702       res = VG_(do_syscall4)(__NR_readlinkat, VKI_AT_FDCWD,
 703                              (UWord)"/proc/self",
 704                              (UWord)pid, sizeof(pid));
 705 #     else
 706       res = VG_(do_syscall3)(__NR_readlink, (UWord)"/proc/self",
 707                              (UWord)pid, sizeof(pid));
 708 #     endif
 709       if (!sr_isError(res) && sr_Res(res) > 0) {
 710          HChar* s;
 711          pid[sr_Res(res)] = '\0';
 712          res = VG_(mk_SysRes_Success)(  VG_(strtoll10)(pid, &s) );
 713          if (*s != '\0') {
 714             VG_(message)(Vg_DebugMsg,
 715                "Warning: invalid file name linked to by /proc/self: %s\n",
 716                pid);
 717          }
 718       }
 719    }
 720
 721    return sr_Res(res);
 722
 723 #  elif defined(VGO_freebsd)
 724    SysRes res;
 725    long tid;
 726
 727    res = VG_(do_syscall1)(__NR_thr_self, (UWord)&tid);
 728    if (sr_isError(res))
 729       tid = sr_Res(VG_(do_syscall0)(__NR_getpid));
 730    return tid;
 731
 732 #  elif defined(VGO_darwin)
 733    // Darwin's gettid syscall is something else.
 734    // Use Mach thread ports for lwpid instead.
 735    return mach_thread_self();
 736
 737 #  elif defined(VGO_solaris)
 738    SysRes res = VG_(do_syscall0)(__NR_lwp_self);
 739    return sr_Res(res);
 740
 741 #  else
 742 #    error "Unknown OS"
 743 #  endif
 744 }
 745
 746 /* You'd be amazed how many places need to know the current pid. */
 747 Int VG_(getpid) ( void )
 748 {
 749    /* ASSUMES SYSCALL ALWAYS SUCCEEDS */
 750    return sr_Res( VG_(do_syscall0)(__NR_getpid) );
 751 }
 752
 753 Int VG_(getpgrp) ( void )
 754 {
 755    /* ASSUMES SYSCALL ALWAYS SUCCEEDS */
 756 #  if defined(VGP_arm64_linux) || defined(VGP_nanomips_linux)
 757    return sr_Res( VG_(do_syscall1)(__NR_getpgid, 0) );
 758 #  elif defined(VGO_linux) || defined(VGO_darwin) || defined(VGO_freebsd)
 759    return sr_Res( VG_(do_syscall0)(__NR_getpgrp) );
 760 #  elif defined(VGO_solaris)
 761    /* Uses the shared pgrpsys syscall, 0 for the getpgrp variant. */
 762    return sr_Res( VG_(do_syscall1)(__NR_pgrpsys, 0) );
 763 #  else
 764 #    error Unknown OS
 765 #  endif
 766 }
 767
 768 Int VG_(getppid) ( void )
 769 {
 770    /* ASSUMES SYSCALL ALWAYS SUCCEEDS */
 771 #  if defined(VGO_linux) || defined(VGO_darwin) || defined(VGO_freebsd)
 772    return sr_Res( VG_(do_syscall0)(__NR_getppid) );
 773 #  elif defined(VGO_solaris)
 774    /* Uses the shared getpid/getppid syscall, val2 contains a parent pid. */
 775    return sr_ResHI( VG_(do_syscall0)(__NR_getpid) );
 776 #  else
 777 #    error Unknown OS
 778 #  endif
 779 }
 780
 781 Int VG_(geteuid) ( void )
 782 {
 783    /* ASSUMES SYSCALL ALWAYS SUCCEEDS */
 784 #  if defined(VGO_linux) || defined(VGO_darwin) || defined(VGO_freebsd)
 785    {
 786 #     if defined(__NR_geteuid32)
 787       // We use the 32-bit version if it's supported.  Otherwise, IDs greater
 788       // than 65536 cause problems, as bug #151209 showed.
 789       return sr_Res( VG_(do_syscall0)(__NR_geteuid32) );
 790 #     else
 791       return sr_Res( VG_(do_syscall0)(__NR_geteuid) );
 792 #     endif
 793    }
 794 #  elif defined(VGO_solaris)
 795    /* Uses the shared getuid/geteuid syscall, val2 contains the effective
 796       uid. */
 797    return sr_ResHI( VG_(do_syscall0)(__NR_getuid) );
 798 #  else
 799 #    error Unknown OS
 800 #  endif
 801 }
 802
 803 Int VG_(getegid) ( void )
 804 {
 805 #  if defined(VGO_linux) || defined(VGO_darwin) || defined(VGO_freebsd)
 806    /* ASSUMES SYSCALL ALWAYS SUCCEEDS */
 807 #    if defined(__NR_getegid32)
 808    // We use the 32-bit version if it's supported.  Otherwise, IDs greater
 809    // than 65536 cause problems, as bug #151209 showed.
 810    return sr_Res( VG_(do_syscall0)(__NR_getegid32) );
 811 #    else
 812    return sr_Res( VG_(do_syscall0)(__NR_getegid) );
 813 #    endif
 814
 815 #  elif defined(VGO_solaris)
 816    /* Uses the shared getgid/getegid syscall, val2 contains the effective
 817       gid. */
 818    return sr_ResHI( VG_(do_syscall0)(__NR_getgid) );
 819 #  else
 820 #    error Unknown OS
 821 #  endif
 822 }
 823
 824 /* Get supplementary groups into list[0 .. size-1].  Returns the
 825    number of groups written, or -1 if error.  Note that in order to be
 826    portable, the groups are 32-bit unsigned ints regardless of the
 827    platform.
 828    As a special case, if size == 0 the function returns the number of
 829    groups leaving list untouched. */
 830 Int VG_(getgroups)( Int size, UInt* list )
 831 {
 832    if (size < 0) return -1;
 833
 834 #  if defined(VGP_x86_linux) || defined(VGP_ppc32_linux) \
 835       || defined(VGP_mips64_linux)
 836    Int    i;
 837    SysRes sres;
 838    UShort list16[size];
 839    sres = VG_(do_syscall2)(__NR_getgroups, size, (Addr)list16);
 840    if (sr_isError(sres))
 841       return -1;
 842    if (size != 0) {
 843       for (i = 0; i < sr_Res(sres); i++)
 844          list[i] = (UInt)list16[i];
 845    }
 846    return sr_Res(sres);
 847
 848 #  elif defined(VGP_amd64_linux) || defined(VGP_arm_linux) \
 849         || defined(VGP_ppc64be_linux) || defined(VGP_ppc64le_linux)  \
 850         || defined(VGO_darwin) || defined(VGP_s390x_linux)    \
 851         || defined(VGP_mips32_linux) || defined(VGP_arm64_linux) \
 852         || defined(VGO_solaris) || defined(VGP_nanomips_linux) \
 853         || defined(VGO_freebsd)
 854    SysRes sres;
 855    sres = VG_(do_syscall2)(__NR_getgroups, size, (Addr)list);
 856    if (sr_isError(sres))
 857       return -1;
 858    return sr_Res(sres);
 859
 860 #  else
 861 #     error "VG_(getgroups): needs implementation on this platform"
 862 #  endif
 863 }
 864
 865 /* ---------------------------------------------------------------------
 866    Process tracing
 867    ------------------------------------------------------------------ */
 868
 869 Int VG_(ptrace) ( Int request, Int pid, void *addr, void *data )
 870 {
 871    SysRes res;
 872 #  if defined(VGO_linux) || defined(VGO_darwin) || defined(VGO_freebsd)
 873    res = VG_(do_syscall4)(__NR_ptrace, request, pid, (UWord)addr, (UWord)data);
 874 #  elif defined(VGO_solaris)
 875    /* There is no ptrace syscall on Solaris.  Such requests has to be
 876       implemented using the /proc interface.  Callers of VG_(ptrace) should
 877       ensure that this function is not reached on Solaris, i.e. they must
 878       provide a special code for Solaris for whatever feature they provide. */
 879    I_die_here;
 880 #  else
 881 #    error Unknown OS
 882 #  endif
 883    if (sr_isError(res))
 884       return -1;
 885    return sr_Res(res);
 886 }
 887
 888 /* ---------------------------------------------------------------------
 889    Fork
 890    ------------------------------------------------------------------ */
 891
 892 /* Record PID of a child process in order to avoid sending any SIGCHLD from
 893    it to the client.  If PID is 0 then this is the child process and it
 894    should synch with the parent to ensure it can't send any SIGCHLD before
 895    the parent has registered its PID.
 896
 897    FDS should be initialized with VG_(pipe). This function closes both
 898    file descriptors.  */
 899 static void register_sigchld_ignore ( Int pid, Int fds[2])
 900 {
 901    Int child_wait = 1;
 902    ht_ignore_node *n;
 903
 904    if (fds[0] < 0 || fds[1] < 0)
 905       return;
 906
 907    if (pid == 0) {
 908       /* We are the child, close writing fd that we don't use.  */
 909       VG_(close)(fds[1]);
 910       /* Before proceeding, ensure parent has recorded child PID in map
 911          of SIGCHLD to ignore */
 912       while (child_wait == 1)
 913       {
 914          if (VG_(read)(fds[0], &child_wait, sizeof(Int)) <= 0) {
 915             VG_(message)(Vg_DebugMsg,
 916                "warning: Unable to record PID of internal process (read)\n");
 917             child_wait = 0;
 918          }
 919       }
 920
 921       /* Now close reading fd.  */
 922       VG_(close)(fds[0]);
 923       return;
 924    }
 925
 926    n = VG_(malloc)("ht.ignore.node", sizeof(ht_ignore_node));
 927    n->key = pid;
 928    if (ht_sigchld_ignore == NULL)
 929       ht_sigchld_ignore = VG_(HT_construct)("ht.sigchld.ignore");
 930    VG_(HT_add_node)(ht_sigchld_ignore, n);
 931
 932    /* We are the parent process, close read fd that we don't use.  */
 933    VG_(close)(fds[0]);
 934
 935    child_wait = 0;
 936    if (VG_(write)(fds[1], &child_wait, sizeof(Int)) <= 0)
 937       VG_(message)(Vg_DebugMsg,
 938          "warning: Unable to record PID of internal process (write)\n");
 939
 940    /* Now close writing fd.  */
 941    VG_(close)(fds[1]);
 942 }
 943
 944 Int VG_(fork) ( void )
 945 {
 946    Int fds[2];
 947
 948    if (VG_(pipe)(fds) != 0) {
 949       VG_(message)(Vg_DebugMsg,
 950          "warning: Unable to record PID of internal process (pipe)\n");
 951       fds[0] = fds[1] = -1;
 952    }
 953
 954 #  if defined(VGP_arm64_linux) || defined(VGP_nanomips_linux)
 955    SysRes res;
 956    res = VG_(do_syscall5)(__NR_clone, VKI_SIGCHLD,
 957                           (UWord)NULL, (UWord)NULL, (UWord)NULL, (UWord)NULL);
 958    if (sr_isError(res))
 959       return -1;
 960    register_sigchld_ignore(sr_Res(res), fds);
 961    return sr_Res(res);
 962
 963 #  elif defined(VGO_linux) || defined(VGO_freebsd)
 964    SysRes res;
 965    res = VG_(do_syscall0)(__NR_fork);
 966    if (sr_isError(res))
 967       return -1;
 968    register_sigchld_ignore(sr_Res(res), fds);
 969    return sr_Res(res);
 970
 971 #  elif defined(VGO_darwin)
 972    SysRes res;
 973    res = VG_(do_syscall0)(__NR_fork); /* __NR_fork is UX64 */
 974    if (sr_isError(res))
 975       return -1;
 976    /* on success: wLO = child pid; wHI = 1 for child, 0 for parent */
 977    if (sr_ResHI(res) != 0) {
 978       register_sigchld_ignore(0, fds);
 979       return 0;  /* this is child: return 0 instead of child pid */
 980    }
 981    register_sigchld_ignore(sr_Res(res), fds);
 982    return sr_Res(res);
 983
 984 #  elif defined(VGO_solaris)
 985    /* Using fork() on Solaris is not really the best thing to do. Solaris
 986       does not do memory overcommitment so fork() can fail if there is not
 987       enough memory to copy the current process into a new one.
 988       Prefer to use VG_(spawn)() over VG_(fork)() + VG_(execv)(). */
 989    SysRes res;
 990    res = VG_(do_syscall2)(__NR_forksys, 0 /*subcode (fork)*/, 0 /*flags*/);
 991    if (sr_isError(res))
 992       return -1;
 993    /* On success:
 994         val = a pid of the child in the parent, a pid of the parent in the
 995               child,
 996         val2 = 0 in the parent process, 1 in the child process. */
 997    if (sr_ResHI(res) != 0) {
 998       register_sigchld_ignore(0, fds);
 999       return 0;
1000    }
1001    register_sigchld_ignore(sr_Res(res), fds);
1002    return sr_Res(res);
1003
1004 #  else
1005 #    error "Unknown OS"
1006 #  endif
1007 }
1008
1009 /* ---------------------------------------------------------------------
1010    Timing stuff
1011    ------------------------------------------------------------------ */
1012
1013 UInt VG_(read_millisecond_timer) ( void )
1014 {
1015    /* 'now' and 'base' are in microseconds */
1016    static ULong base = 0;
1017    ULong  now;
1018
1019 #  if defined(VGO_linux) || defined(VGO_solaris)
1020    { SysRes res;
1021      struct vki_timespec ts_now;
1022      res = VG_(do_syscall2)(__NR_clock_gettime, VKI_CLOCK_MONOTONIC,
1023                             (UWord)&ts_now);
1024      if (sr_isError(res) == 0) {
1025         now = ts_now.tv_sec * 1000000ULL + ts_now.tv_nsec / 1000;
1026      } else {
1027        struct vki_timeval tv_now;
1028        /* Note: On Solaris, this syscall takes only one parameter but the
1029           extra dummy one does not cause any harm. */
1030        res = VG_(do_syscall2)(__NR_gettimeofday, (UWord)&tv_now, (UWord)NULL);
1031        vg_assert(! sr_isError(res));
1032        now = tv_now.tv_sec * 1000000ULL + tv_now.tv_usec;
1033      }
1034    }
1035
1036 #  elif defined(VGO_freebsd)
1037    { SysRes res;
1038      struct vki_timeval tv_now;
1039      res = VG_(do_syscall2)(__NR_gettimeofday, (UWord)&tv_now, (UWord)NULL);
1040      vg_assert(! sr_isError(res));
1041      now = tv_now.tv_sec * 1000000ULL + tv_now.tv_usec;
1042    }
1043 #  elif defined(VGO_darwin)
1044    // Weird: it seems that gettimeofday() doesn't fill in the timeval, but
1045    // rather returns the tv_sec as the low 32 bits of the result and the
1046    // tv_usec as the high 32 bits of the result.  (But the timeval cannot be
1047    // NULL!)  See bug 200990.
1048    { SysRes res;
1049      struct vki_timeval tv_now = { 0, 0 };
1050      res = VG_(do_syscall2)(__NR_gettimeofday, (UWord)&tv_now, (UWord)NULL);
1051      vg_assert(! sr_isError(res));
1052      now = sr_Res(res) * 1000000ULL + sr_ResHI(res);
1053    }
1054
1055 #  else
1056 #    error "Unknown OS"
1057 #  endif
1058
1059    /* COMMON CODE */
1060    if (base == 0)
1061       base = now;
1062
1063    return (now - base) / 1000;
1064 }
1065
1066 #  if defined(VGO_linux) || defined(VGO_solaris) || defined(VGO_freebsd)
1067 void VG_(clock_gettime) ( struct vki_timespec *ts, vki_clockid_t clk_id )
1068 {
1069     SysRes res;
1070     res = VG_(do_syscall2)(__NR_clock_gettime, clk_id,
1071                            (UWord)ts);
1072     vg_assert (sr_isError(res) == 0);
1073 }
1074 #  elif defined(VGO_darwin)
1075   /* See pub_tool_libcproc.h */
1076 #  else
1077 #    error "Unknown OS"
1078 #  endif
1079
1080 Int VG_(gettimeofday)(struct vki_timeval *tv, struct vki_timezone *tz)
1081 {
1082    SysRes res;
1083    res = VG_(do_syscall2)(__NR_gettimeofday, (UWord)tv, (UWord)tz);
1084
1085    if (! sr_isError(res)) return 0;
1086
1087    /* Make sure, argument values are deterministic upon failure */
1088    if (tv) *tv = (struct vki_timeval){ .tv_sec = 0, .tv_usec = 0 };
1089    if (tz) *tz = (struct vki_timezone){ .tz_minuteswest = 0, .tz_dsttime = 0 };
1090
1091    return -1;
1092 }
1093
1094 UInt VG_(get_user_milliseconds)(void)
1095 {
1096    UInt res = 0;
1097 #  if defined(VGO_linux)
1098    {
1099       struct vki_rusage ru;
1100       VG_(memset)(&ru, 0, sizeof(ru));
1101       SysRes sr = VG_(do_syscall2)(__NR_getrusage, VKI_RUSAGE_SELF, (UWord)&ru);
1102       if (!sr_isError(sr)) {
1103          res = ru.ru_utime.tv_sec * 1000 + ru.ru_utime.tv_usec / 1000;
1104       }
1105    }
1106
1107 #  elif defined(VGO_solaris)
1108    {
1109       struct vki_rusage ru;
1110       VG_(memset)(&ru, 0, sizeof(ru));
1111       SysRes sr = VG_(do_syscall2)(__NR_rusagesys, VKI__RUSAGESYS_GETRUSAGE,
1112                                    (UWord) &ru);
1113       if (!sr_isError(sr)) {
1114          res = ru.ru_utime.tv_sec * 1000 + ru.ru_utime.tv_usec / 1000;
1115       }
1116    }
1117
1118 #  elif defined(VGO_freebsd)
1119    {
1120       struct vki_rusage ru;
1121       VG_(memset)(&ru, 0, sizeof(ru));
1122       SysRes sr = VG_(do_syscall2)(__NR_getrusage, VKI_RUSAGE_SELF, (UWord)&ru);
1123       if (!sr_isError(sr)) {
1124          res = ru.ru_utime.tv_sec * 1000 + ru.ru_utime.tv_usec / 1000;
1125       }
1126    }
1127
1128 #  elif defined(VGO_darwin)
1129    res = 0;
1130
1131 #  else
1132 #    error "Unknown OS"
1133 #  endif
1134
1135    return res;
1136 }
1137
1138
1139 /* ---------------------------------------------------------------------
1140    atfork()
1141    ------------------------------------------------------------------ */
1142
1143 struct atfork {
1144    vg_atfork_t  pre;
1145    vg_atfork_t  parent;
1146    vg_atfork_t  child;
1147 };
1148
1149 #define VG_MAX_ATFORK 10
1150
1151 static struct atfork atforks[VG_MAX_ATFORK];
1152 static Int n_atfork = 0;
1153
1154 void VG_(atfork)(vg_atfork_t pre, vg_atfork_t parent, vg_atfork_t child)
1155 {
1156    Int i;
1157
1158    for (i = 0; i < n_atfork; i++) {
1159       if (atforks[i].pre == pre &&
1160           atforks[i].parent == parent &&
1161           atforks[i].child == child)
1162          return;
1163    }
1164
1165    if (n_atfork >= VG_MAX_ATFORK)
1166       VG_(core_panic)(
1167          "Too many VG_(atfork) handlers requested: raise VG_MAX_ATFORK");
1168
1169    atforks[n_atfork].pre    = pre;
1170    atforks[n_atfork].parent = parent;
1171    atforks[n_atfork].child  = child;
1172
1173    n_atfork++;
1174 }
1175
1176 void VG_(do_atfork_pre)(ThreadId tid)
1177 {
1178    Int i;
1179
1180    for (i = 0; i < n_atfork; i++)
1181       if (atforks[i].pre != NULL)
1182          (*atforks[i].pre)(tid);
1183 }
1184
1185 void VG_(do_atfork_parent)(ThreadId tid)
1186 {
1187    Int i;
1188
1189    for (i = 0; i < n_atfork; i++)
1190       if (atforks[i].parent != NULL)
1191          (*atforks[i].parent)(tid);
1192 }
1193
1194 void VG_(do_atfork_child)(ThreadId tid)
1195 {
1196    Int i;
1197
1198    for (i = 0; i < n_atfork; i++)
1199       if (atforks[i].child != NULL)
1200          (*atforks[i].child)(tid);
1201 }
1202
1203 /* ---------------------------------------------------------------------
1204    FreeBSD sysctlbyname, getosreldate, is32on64
1205    ------------------------------------------------------------------ */
1206
1207 #if defined(VGO_freebsd)
1208 Int VG_(sysctlbyname)(const HChar *name, void *oldp, SizeT *oldlenp, const void *newp, SizeT newlen)
1209 {
1210    vg_assert(name);
1211 #if (FREEBSD_VERS >= FREEBSD_12_2)
1212    SysRes res = VG_(do_syscall6)(__NR___sysctlbyname, (RegWord)name, VG_(strlen)(name), (RegWord)oldp, (RegWord)oldlenp, (RegWord)newp, (RegWord)newlen);
1213    return sr_isError(res) ? -1 : sr_Res(res);
1214 #else
1215    Int oid[2];
1216    Int real_oid[10];
1217    SizeT oidlen;
1218    int error;
1219
1220    oid[0] = 0;          /* magic */
1221    oid[1] = 3;          /* undocumented */
1222    oidlen = sizeof(real_oid);
1223    error = VG_(sysctl)(oid, 2, real_oid, &oidlen, name, VG_(strlen)(name));
1224    if (error < 0)
1225       return error;
1226    oidlen /= sizeof(int);
1227    error = VG_(sysctl)(real_oid, oidlen, oldp, oldlenp, newp, newlen);
1228    return error;
1229  #endif
1230 }
1231
1232 Int VG_(getosreldate)(void)
1233 {
1234    static Int osreldate = 0;
1235    SizeT osreldatel;
1236
1237    if (osreldate == 0) {
1238       osreldatel = sizeof(osreldate);
1239       VG_(sysctlbyname)("kern.osreldate", &osreldate, &osreldatel, 0, 0);
1240    }
1241    return (osreldate);
1242 }
1243
1244 Bool VG_(is32on64)(void)
1245 {
1246 #if defined(VGP_amd64_freebsd) || defined(VGP_arm64_freebsd)
1247    return False;
1248 #elif defined(VGP_x86_freebsd)
1249    SysRes res;
1250    struct vg_stat stat_buf;
1251    res = VG_(stat)("/libexec/ld-elf32.so.1", &stat_buf);
1252    if (!sr_isError(res)) {
1253       // file exists, we're running on amd64
1254       VG_(debugLog)(1, "check-os-bitness", "i386 executable on amd64 kernel\n");
1255       return True;
1256    } else {
1257       VG_(debugLog)(1, "check-os-bitness", "i386 executable on i386 kernel\n");
1258       return False;
1259    }
1260 #else
1261 #  error Unknown platform
1262 #endif
1263 }
1264 #endif
1265
1266 /* ---------------------------------------------------------------------
1267    icache invalidation
1268    ------------------------------------------------------------------ */
1269
1270 void VG_(invalidate_icache) ( void *ptr, SizeT nbytes )
1271 {
1272    if (nbytes == 0) return;    // nothing to do
1273
1274    // Get cache info
1275    VexArchInfo vai;
1276    VG_(machine_get_VexArchInfo)(NULL, &vai);
1277
1278    // If I-caches are coherent, nothing needs to be done here
1279    if (vai.hwcache_info.icaches_maintain_coherence) return;
1280
1281 #  if defined(VGA_ppc32) || defined(VGA_ppc64be) || defined(VGA_ppc64le)
1282    Addr startaddr = (Addr) ptr;
1283    Addr endaddr   = startaddr + nbytes;
1284    Addr cls;
1285    Addr addr;
1286
1287    cls = vai.ppc_icache_line_szB;
1288
1289    /* Stay sane .. */
1290    vg_assert(cls == 16 || cls == 32 || cls == 64 || cls == 128);
1291
1292    startaddr &= ~(cls - 1);
1293    for (addr = startaddr; addr < endaddr; addr += cls) {
1294       __asm__ __volatile__("dcbst 0,%0" : : "r" (addr));
1295    }
1296    __asm__ __volatile__("sync");
1297    for (addr = startaddr; addr < endaddr; addr += cls) {
1298       __asm__ __volatile__("icbi 0,%0" : : "r" (addr));
1299    }
1300    __asm__ __volatile__("sync; isync");
1301
1302 #  elif defined(VGP_arm_linux)
1303    /* ARM cache flushes are privileged, so we must defer to the kernel. */
1304    Addr startaddr = (Addr) ptr;
1305    Addr endaddr   = startaddr + nbytes;
1306    VG_(do_syscall2)(__NR_ARM_cacheflush, startaddr, endaddr);
1307
1308 #  elif defined(VGP_arm64_linux) || defined(VGP_arm64_freebsd)
1309    // This arm64_linux section of this function VG_(invalidate_icache)
1310    // is copied from
1311    // https://github.com/armvixl/vixl/blob/master/src/a64/cpu-a64.cc
1312    // which has the following copyright notice:
1313    /*
1314    Copyright 2013, ARM Limited
1315    All rights reserved.
1316
1317    Redistribution and use in source and binary forms, with or without
1318    modification, are permitted provided that the following conditions are met:
1319
1320    * Redistributions of source code must retain the above copyright notice,
1321      this list of conditions and the following disclaimer.
1322    * Redistributions in binary form must reproduce the above copyright notice,
1323      this list of conditions and the following disclaimer in the documentation
1324      and/or other materials provided with the distribution.
1325    * Neither the name of ARM Limited nor the names of its contributors may be
1326      used to endorse or promote products derived from this software without
1327      specific prior written permission.
1328
1329    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS CONTRIBUTORS "AS IS" AND
1330    ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1331    WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1332    DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE
1333    FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
1334    DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1335    SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
1336    CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
1337    OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
1338    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
1339    */
1340
1341    // Ask what the I and D line sizes are
1342    ULong read_mrs;
1343    // Copy the content of the cache type register to a core register.
1344    __asm__ __volatile__ ("mrs %[ctr], ctr_el0" // NOLINT
1345                          : [ctr] "=r" (read_mrs));
1346
1347    UInt cache_type_register = read_mrs;
1348    const Int kDCacheLineSizeShift = 16;
1349    const Int kICacheLineSizeShift = 0;
1350    const UInt kDCacheLineSizeMask = 0xf << kDCacheLineSizeShift;
1351    const UInt kICacheLineSizeMask = 0xf << kICacheLineSizeShift;
1352
1353    // The cache type register holds the size of the I and D caches as a power of
1354    // two.
1355    const UInt dcache_line_size_power_of_two =
1356        (cache_type_register & kDCacheLineSizeMask) >> kDCacheLineSizeShift;
1357    const UInt icache_line_size_power_of_two =
1358        (cache_type_register & kICacheLineSizeMask) >> kICacheLineSizeShift;
1359
1360    const UInt dcache_line_size_ = 4 * (1 << dcache_line_size_power_of_two);
1361    const UInt icache_line_size_ = 4 * (1 << icache_line_size_power_of_two);
1362
1363    Addr start = (Addr)ptr;
1364    // Sizes will be used to generate a mask big enough to cover a pointer.
1365    Addr dsize = (Addr)dcache_line_size_;
1366    Addr isize = (Addr)icache_line_size_;
1367
1368    // Cache line sizes are always a power of 2.
1369    Addr dstart = start & ~(dsize - 1);
1370    Addr istart = start & ~(isize - 1);
1371    Addr end    = start + nbytes;
1372
1373    __asm__ __volatile__ (
1374      // Clean every line of the D cache containing the target data.
1375      "0: \n\t"
1376      // dc : Data Cache maintenance
1377      // c : Clean
1378      // va : by (Virtual) Address
1379      // u : to the point of Unification
1380      // The point of unification for a processor is the point by which the
1381      // instruction and data caches are guaranteed to see the same copy of a
1382      // memory location. See ARM DDI 0406B page B2-12 for more information.
1383      "dc cvau, %[dline] \n\t"
1384      "add %[dline], %[dline], %[dsize] \n\t"
1385      "cmp %[dline], %[end] \n\t"
1386      "b.lt 0b \n\t"
1387      // Barrier to make sure the effect of the code above is visible to the rest
1388      // of the world.
1389      // dsb : Data Synchronisation Barrier
1390      // ish : Inner SHareable domain
1391      // The point of unification for an Inner Shareable shareability domain is
1392      // the point by which the instruction and data caches of all the processors
1393      // in that Inner Shareable shareability domain are guaranteed to see the
1394      // same copy of a memory location. See ARM DDI 0406B page B2-12 for more
1395      // information.
1396      "dsb ish \n\t"
1397      // Invalidate every line of the I cache containing the target data.
1398      "1: \n\t"
1399      // ic : instruction cache maintenance
1400      // i : invalidate
1401      // va : by address
1402      // u : to the point of unification
1403      "ic ivau, %[iline] \n\t"
1404      "add %[iline], %[iline], %[isize] \n\t"
1405      "cmp %[iline], %[end] \n\t"
1406      "b.lt 1b \n\t"
1407      // Barrier to make sure the effect of the code above is visible to the rest
1408      // of the world.
1409      "dsb ish \n\t"
1410      // Barrier to ensure any prefetching which happened before this code is
1411      // discarded.
1412      // isb : Instruction Synchronisation Barrier
1413      "isb \n\t"
1414      : [dline] "+r" (dstart),
1415        [iline] "+r" (istart)
1416      : [dsize] "r" (dsize),
1417        [isize] "r" (isize),
1418        [end] "r" (end)
1419      // This code does not write to memory but without the dependency gcc might
1420      // move this code before the code is generated.
1421      : "cc", "memory"
1422    );
1423
1424 #  elif defined(VGA_mips32) || defined(VGA_mips64)
1425    SysRes sres = VG_(do_syscall3)(__NR_cacheflush, (UWord) ptr,
1426                                  (UWord) nbytes, (UWord) 3);
1427    vg_assert( !sr_isError(sres) );
1428
1429 # elif defined(VGA_nanomips)
1430
1431    __builtin___clear_cache(ptr, (char*)ptr + nbytes);
1432
1433 #  endif
1434 }
1435
1436
1437 /* ---------------------------------------------------------------------
1438    dcache flushing
1439    ------------------------------------------------------------------ */
1440
1441 void VG_(flush_dcache) ( void *ptr, SizeT nbytes )
1442 {
1443    /* Currently this is only required on ARM64. */
1444 #  if defined(VGA_arm64)
1445    Addr startaddr = (Addr) ptr;
1446    Addr endaddr   = startaddr + nbytes;
1447    Addr cls;
1448    Addr addr;
1449
1450    ULong ctr_el0;
1451    __asm__ __volatile__ ("mrs %0, ctr_el0" : "=r"(ctr_el0));
1452    cls = 4 * (1ULL << (0xF & (ctr_el0 >> 16)));
1453
1454    /* Stay sane .. */
1455    vg_assert(cls == 64 || cls == 128);
1456
1457    startaddr &= ~(cls - 1);
1458    for (addr = startaddr; addr < endaddr; addr += cls) {
1459       __asm__ __volatile__("dc cvau, %0" : : "r" (addr));
1460    }
1461    __asm__ __volatile__("dsb ish");
1462 #  endif
1463 }
1464
1465 /*--------------------------------------------------------------------*/
1466 /*--- end                                                          ---*/
1467 /*--------------------------------------------------------------------*/