• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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, write to the Free Software
25    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
26    02111-1307, USA.
27 
28    The GNU General Public License is contained in the file COPYING.
29 */
30 
31 #include "pub_core_basics.h"
32 #include "pub_core_machine.h"    // For VG_(machine_get_VexArchInfo)
33 #include "pub_core_vki.h"
34 #include "pub_core_vkiscnums.h"
35 #include "pub_core_libcbase.h"
36 #include "pub_core_libcassert.h"
37 #include "pub_core_libcfile.h"
38 #include "pub_core_libcprint.h"
39 #include "pub_core_libcproc.h"
40 #include "pub_core_libcsignal.h"
41 #include "pub_core_seqmatch.h"
42 #include "pub_core_mallocfree.h"
43 #include "pub_core_syscall.h"
44 #include "pub_core_xarray.h"
45 #include "pub_core_clientstate.h"
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)
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. */
VG_(getenv)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. */
VG_(env_unsetenv)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 */
VG_(env_setenv)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 */
mash_colon_env(HChar * varp,const HChar * remove_pattern)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. */
VG_(env_remove_valgrind_env_stuff)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'. */
VG_(client_fname)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 
add_string(HChar * buffer,SizeT * buf_size,const HChar * string)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'. */
VG_(client_cmd_and_args)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 
VG_(waitpid)349 Int VG_(waitpid)(Int pid, Int *status, Int options)
350 {
351 #  if defined(VGO_linux)
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 */
VG_(env_clone)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 
VG_(execv)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. */
VG_(spawn)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! */
VG_(system)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 
VG_(sysctl)586 Int VG_(sysctl)(Int *name, UInt namelen, void *oldp, SizeT *oldlenp, void *newp, SizeT newlen)
587 {
588    SysRes res;
589 #  if defined(VGO_darwin)
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. */
VG_(getrlimit)603 Int VG_(getrlimit) (Int resource, struct vki_rlimit *rlim)
604 {
605    SysRes res = VG_(mk_SysRes_Error)(VKI_ENOSYS);
606    /* res = getrlimit( resource, rlim ); */
607 #  ifdef __NR_ugetrlimit
608    res = VG_(do_syscall2)(__NR_ugetrlimit, resource, (UWord)rlim);
609 #  endif
610    if (sr_isError(res) && sr_Err(res) == VKI_ENOSYS)
611       res = VG_(do_syscall2)(__NR_getrlimit, resource, (UWord)rlim);
612    return sr_isError(res) ? -1 : sr_Res(res);
613 }
614 
615 
616 /* Support for setrlimit. */
VG_(setrlimit)617 Int VG_(setrlimit) (Int resource, const struct vki_rlimit *rlim)
618 {
619    SysRes res;
620    /* res = setrlimit( resource, rlim ); */
621    res = VG_(do_syscall2)(__NR_setrlimit, resource, (UWord)rlim);
622    return sr_isError(res) ? -1 : sr_Res(res);
623 }
624 
625 /* Support for prctl. */
VG_(prctl)626 Int VG_(prctl) (Int option,
627                 ULong arg2, ULong arg3, ULong arg4, ULong arg5)
628 {
629    SysRes res = VG_(mk_SysRes_Error)(VKI_ENOSYS);
630 #  if defined(VGO_linux)
631    /* res = prctl( option, arg2, arg3, arg4, arg5 ); */
632    res = VG_(do_syscall5)(__NR_prctl, (UWord) option,
633                           (UWord) arg2, (UWord) arg3, (UWord) arg4,
634                           (UWord) arg5);
635 #  endif
636 
637    return sr_isError(res) ? -1 : sr_Res(res);
638 }
639 
640 /* ---------------------------------------------------------------------
641    pids, etc
642    ------------------------------------------------------------------ */
643 
VG_(gettid)644 Int VG_(gettid)(void)
645 {
646 #  if defined(VGO_linux)
647    SysRes res = VG_(do_syscall0)(__NR_gettid);
648 
649    if (sr_isError(res) && sr_Res(res) == VKI_ENOSYS) {
650       HChar pid[16];
651       /*
652        * The gettid system call does not exist. The obvious assumption
653        * to make at this point would be that we are running on an older
654        * system where the getpid system call actually returns the ID of
655        * the current thread.
656        *
657        * Unfortunately it seems that there are some systems with a kernel
658        * where getpid has been changed to return the ID of the thread group
659        * leader but where the gettid system call has not yet been added.
660        *
661        * So instead of calling getpid here we use readlink to see where
662        * the /proc/self link is pointing...
663        */
664 
665 #     if defined(VGP_arm64_linux)
666       res = VG_(do_syscall4)(__NR_readlinkat, VKI_AT_FDCWD,
667                              (UWord)"/proc/self",
668                              (UWord)pid, sizeof(pid));
669 #     else
670       res = VG_(do_syscall3)(__NR_readlink, (UWord)"/proc/self",
671                              (UWord)pid, sizeof(pid));
672 #     endif
673       if (!sr_isError(res) && sr_Res(res) > 0) {
674          HChar* s;
675          pid[sr_Res(res)] = '\0';
676          res = VG_(mk_SysRes_Success)(  VG_(strtoll10)(pid, &s) );
677          if (*s != '\0') {
678             VG_(message)(Vg_DebugMsg,
679                "Warning: invalid file name linked to by /proc/self: %s\n",
680                pid);
681          }
682       }
683    }
684 
685    return sr_Res(res);
686 
687 #  elif defined(VGO_darwin)
688    // Darwin's gettid syscall is something else.
689    // Use Mach thread ports for lwpid instead.
690    return mach_thread_self();
691 
692 #  elif defined(VGO_solaris)
693    SysRes res = VG_(do_syscall0)(__NR_lwp_self);
694    return sr_Res(res);
695 
696 #  else
697 #    error "Unknown OS"
698 #  endif
699 }
700 
701 /* You'd be amazed how many places need to know the current pid. */
VG_(getpid)702 Int VG_(getpid) ( void )
703 {
704    /* ASSUMES SYSCALL ALWAYS SUCCEEDS */
705    return sr_Res( VG_(do_syscall0)(__NR_getpid) );
706 }
707 
VG_(getpgrp)708 Int VG_(getpgrp) ( void )
709 {
710    /* ASSUMES SYSCALL ALWAYS SUCCEEDS */
711 #  if defined(VGP_arm64_linux)
712    return sr_Res( VG_(do_syscall1)(__NR_getpgid, 0) );
713 #  elif defined(VGO_linux) || defined(VGO_darwin)
714    return sr_Res( VG_(do_syscall0)(__NR_getpgrp) );
715 #  elif defined(VGO_solaris)
716    /* Uses the shared pgrpsys syscall, 0 for the getpgrp variant. */
717    return sr_Res( VG_(do_syscall1)(__NR_pgrpsys, 0) );
718 #  else
719 #    error Unknown OS
720 #  endif
721 }
722 
VG_(getppid)723 Int VG_(getppid) ( void )
724 {
725    /* ASSUMES SYSCALL ALWAYS SUCCEEDS */
726 #  if defined(VGO_linux) || defined(VGO_darwin)
727    return sr_Res( VG_(do_syscall0)(__NR_getppid) );
728 #  elif defined(VGO_solaris)
729    /* Uses the shared getpid/getppid syscall, val2 contains a parent pid. */
730    return sr_ResHI( VG_(do_syscall0)(__NR_getpid) );
731 #  else
732 #    error Unknown OS
733 #  endif
734 }
735 
VG_(geteuid)736 Int VG_(geteuid) ( void )
737 {
738    /* ASSUMES SYSCALL ALWAYS SUCCEEDS */
739 #  if defined(VGO_linux) || defined(VGO_darwin)
740    {
741 #     if defined(__NR_geteuid32)
742       // We use the 32-bit version if it's supported.  Otherwise, IDs greater
743       // than 65536 cause problems, as bug #151209 showed.
744       return sr_Res( VG_(do_syscall0)(__NR_geteuid32) );
745 #     else
746       return sr_Res( VG_(do_syscall0)(__NR_geteuid) );
747 #     endif
748    }
749 #  elif defined(VGO_solaris)
750    /* Uses the shared getuid/geteuid syscall, val2 contains the effective
751       uid. */
752    return sr_ResHI( VG_(do_syscall0)(__NR_getuid) );
753 #  else
754 #    error Unknown OS
755 #  endif
756 }
757 
VG_(getegid)758 Int VG_(getegid) ( void )
759 {
760 #  if defined(VGO_linux) || defined(VGO_darwin)
761    /* ASSUMES SYSCALL ALWAYS SUCCEEDS */
762 #    if defined(__NR_getegid32)
763    // We use the 32-bit version if it's supported.  Otherwise, IDs greater
764    // than 65536 cause problems, as bug #151209 showed.
765    return sr_Res( VG_(do_syscall0)(__NR_getegid32) );
766 #    else
767    return sr_Res( VG_(do_syscall0)(__NR_getegid) );
768 #    endif
769 
770 #  elif defined(VGO_solaris)
771    /* Uses the shared getgid/getegid syscall, val2 contains the effective
772       gid. */
773    return sr_ResHI( VG_(do_syscall0)(__NR_getgid) );
774 #  else
775 #    error Unknown OS
776 #  endif
777 }
778 
779 /* Get supplementary groups into list[0 .. size-1].  Returns the
780    number of groups written, or -1 if error.  Note that in order to be
781    portable, the groups are 32-bit unsigned ints regardless of the
782    platform.
783    As a special case, if size == 0 the function returns the number of
784    groups leaving list untouched. */
VG_(getgroups)785 Int VG_(getgroups)( Int size, UInt* list )
786 {
787    if (size < 0) return -1;
788 
789 #  if defined(VGP_x86_linux) || defined(VGP_ppc32_linux) \
790       || defined(VGP_mips64_linux)
791    Int    i;
792    SysRes sres;
793    UShort list16[size];
794    sres = VG_(do_syscall2)(__NR_getgroups, size, (Addr)list16);
795    if (sr_isError(sres))
796       return -1;
797    if (size != 0) {
798       for (i = 0; i < sr_Res(sres); i++)
799          list[i] = (UInt)list16[i];
800    }
801    return sr_Res(sres);
802 
803 #  elif defined(VGP_amd64_linux) || defined(VGP_arm_linux) \
804         || defined(VGP_ppc64be_linux) || defined(VGP_ppc64le_linux)  \
805         || defined(VGO_darwin) || defined(VGP_s390x_linux)    \
806         || defined(VGP_mips32_linux) || defined(VGP_arm64_linux) \
807         || defined(VGO_solaris)
808    SysRes sres;
809    sres = VG_(do_syscall2)(__NR_getgroups, size, (Addr)list);
810    if (sr_isError(sres))
811       return -1;
812    return sr_Res(sres);
813 
814 #  else
815 #     error "VG_(getgroups): needs implementation on this platform"
816 #  endif
817 }
818 
819 /* ---------------------------------------------------------------------
820    Process tracing
821    ------------------------------------------------------------------ */
822 
VG_(ptrace)823 Int VG_(ptrace) ( Int request, Int pid, void *addr, void *data )
824 {
825    SysRes res;
826 #  if defined(VGO_linux) || defined(VGO_darwin)
827    res = VG_(do_syscall4)(__NR_ptrace, request, pid, (UWord)addr, (UWord)data);
828 #  elif defined(VGO_solaris)
829    /* There is no ptrace syscall on Solaris.  Such requests has to be
830       implemented using the /proc interface.  Callers of VG_(ptrace) should
831       ensure that this function is not reached on Solaris, i.e. they must
832       provide a special code for Solaris for whatever feature they provide. */
833    I_die_here;
834 #  else
835 #    error Unknown OS
836 #  endif
837    if (sr_isError(res))
838       return -1;
839    return sr_Res(res);
840 }
841 
842 /* ---------------------------------------------------------------------
843    Fork
844    ------------------------------------------------------------------ */
845 
VG_(fork)846 Int VG_(fork) ( void )
847 {
848 #  if defined(VGP_arm64_linux)
849    SysRes res;
850    res = VG_(do_syscall5)(__NR_clone, VKI_SIGCHLD,
851                           (UWord)NULL, (UWord)NULL, (UWord)NULL, (UWord)NULL);
852    if (sr_isError(res))
853       return -1;
854    return sr_Res(res);
855 
856 #  elif defined(VGO_linux)
857    SysRes res;
858    res = VG_(do_syscall0)(__NR_fork);
859    if (sr_isError(res))
860       return -1;
861    return sr_Res(res);
862 
863 #  elif defined(VGO_darwin)
864    SysRes res;
865    res = VG_(do_syscall0)(__NR_fork); /* __NR_fork is UX64 */
866    if (sr_isError(res))
867       return -1;
868    /* on success: wLO = child pid; wHI = 1 for child, 0 for parent */
869    if (sr_ResHI(res) != 0) {
870       return 0;  /* this is child: return 0 instead of child pid */
871    }
872    return sr_Res(res);
873 
874 #  elif defined(VGO_solaris)
875    /* Using fork() on Solaris is not really the best thing to do. Solaris
876       does not do memory overcommitment so fork() can fail if there is not
877       enough memory to copy the current process into a new one.
878       Prefer to use VG_(spawn)() over VG_(fork)() + VG_(execv)(). */
879    SysRes res;
880    res = VG_(do_syscall2)(__NR_forksys, 0 /*subcode (fork)*/, 0 /*flags*/);
881    if (sr_isError(res))
882       return -1;
883    /* On success:
884         val = a pid of the child in the parent, a pid of the parent in the
885               child,
886         val2 = 0 in the parent process, 1 in the child process. */
887    if (sr_ResHI(res) != 0) {
888       return 0;
889    }
890    return sr_Res(res);
891 
892 #  else
893 #    error "Unknown OS"
894 #  endif
895 }
896 
897 /* ---------------------------------------------------------------------
898    Timing stuff
899    ------------------------------------------------------------------ */
900 
VG_(read_millisecond_timer)901 UInt VG_(read_millisecond_timer) ( void )
902 {
903    /* 'now' and 'base' are in microseconds */
904    static ULong base = 0;
905    ULong  now;
906 
907 #  if defined(VGO_linux) || defined(VGO_solaris)
908    { SysRes res;
909      struct vki_timespec ts_now;
910      res = VG_(do_syscall2)(__NR_clock_gettime, VKI_CLOCK_MONOTONIC,
911                             (UWord)&ts_now);
912      if (sr_isError(res) == 0) {
913         now = ts_now.tv_sec * 1000000ULL + ts_now.tv_nsec / 1000;
914      } else {
915        struct vki_timeval tv_now;
916        /* Note: On Solaris, this syscall takes only one parameter but the
917           extra dummy one does not cause any harm. */
918        res = VG_(do_syscall2)(__NR_gettimeofday, (UWord)&tv_now, (UWord)NULL);
919        vg_assert(! sr_isError(res));
920        now = tv_now.tv_sec * 1000000ULL + tv_now.tv_usec;
921      }
922    }
923 
924 #  elif defined(VGO_darwin)
925    // Weird: it seems that gettimeofday() doesn't fill in the timeval, but
926    // rather returns the tv_sec as the low 32 bits of the result and the
927    // tv_usec as the high 32 bits of the result.  (But the timeval cannot be
928    // NULL!)  See bug 200990.
929    { SysRes res;
930      struct vki_timeval tv_now = { 0, 0 };
931      res = VG_(do_syscall2)(__NR_gettimeofday, (UWord)&tv_now, (UWord)NULL);
932      vg_assert(! sr_isError(res));
933      now = sr_Res(res) * 1000000ULL + sr_ResHI(res);
934    }
935 
936 #  else
937 #    error "Unknown OS"
938 #  endif
939 
940    /* COMMON CODE */
941    if (base == 0)
942       base = now;
943 
944    return (now - base) / 1000;
945 }
946 
VG_(gettimeofday)947 Int VG_(gettimeofday)(struct vki_timeval *tv, struct vki_timezone *tz)
948 {
949    SysRes res;
950    res = VG_(do_syscall2)(__NR_gettimeofday, (UWord)tv, (UWord)tz);
951 
952    if (! sr_isError(res)) return 0;
953 
954    /* Make sure, argument values are deterministic upon failure */
955    if (tv) *tv = (struct vki_timeval){ .tv_sec = 0, .tv_usec = 0 };
956    if (tz) *tz = (struct vki_timezone){ .tz_minuteswest = 0, .tz_dsttime = 0 };
957 
958    return -1;
959 }
960 
961 
962 /* ---------------------------------------------------------------------
963    atfork()
964    ------------------------------------------------------------------ */
965 
966 struct atfork {
967    vg_atfork_t  pre;
968    vg_atfork_t  parent;
969    vg_atfork_t  child;
970 };
971 
972 #define VG_MAX_ATFORK 10
973 
974 static struct atfork atforks[VG_MAX_ATFORK];
975 static Int n_atfork = 0;
976 
VG_(atfork)977 void VG_(atfork)(vg_atfork_t pre, vg_atfork_t parent, vg_atfork_t child)
978 {
979    Int i;
980 
981    for (i = 0; i < n_atfork; i++) {
982       if (atforks[i].pre == pre &&
983           atforks[i].parent == parent &&
984           atforks[i].child == child)
985          return;
986    }
987 
988    if (n_atfork >= VG_MAX_ATFORK)
989       VG_(core_panic)(
990          "Too many VG_(atfork) handlers requested: raise VG_MAX_ATFORK");
991 
992    atforks[n_atfork].pre    = pre;
993    atforks[n_atfork].parent = parent;
994    atforks[n_atfork].child  = child;
995 
996    n_atfork++;
997 }
998 
VG_(do_atfork_pre)999 void VG_(do_atfork_pre)(ThreadId tid)
1000 {
1001    Int i;
1002 
1003    for (i = 0; i < n_atfork; i++)
1004       if (atforks[i].pre != NULL)
1005          (*atforks[i].pre)(tid);
1006 }
1007 
VG_(do_atfork_parent)1008 void VG_(do_atfork_parent)(ThreadId tid)
1009 {
1010    Int i;
1011 
1012    for (i = 0; i < n_atfork; i++)
1013       if (atforks[i].parent != NULL)
1014          (*atforks[i].parent)(tid);
1015 }
1016 
VG_(do_atfork_child)1017 void VG_(do_atfork_child)(ThreadId tid)
1018 {
1019    Int i;
1020 
1021    for (i = 0; i < n_atfork; i++)
1022       if (atforks[i].child != NULL)
1023          (*atforks[i].child)(tid);
1024 }
1025 
1026 
1027 /* ---------------------------------------------------------------------
1028    icache invalidation
1029    ------------------------------------------------------------------ */
1030 
VG_(invalidate_icache)1031 void VG_(invalidate_icache) ( void *ptr, SizeT nbytes )
1032 {
1033    if (nbytes == 0) return;    // nothing to do
1034 
1035    // Get cache info
1036    VexArchInfo vai;
1037    VG_(machine_get_VexArchInfo)(NULL, &vai);
1038 
1039    // If I-caches are coherent, nothing needs to be done here
1040    if (vai.hwcache_info.icaches_maintain_coherence) return;
1041 
1042 #  if defined(VGA_ppc32) || defined(VGA_ppc64be) || defined(VGA_ppc64le)
1043    Addr startaddr = (Addr) ptr;
1044    Addr endaddr   = startaddr + nbytes;
1045    Addr cls;
1046    Addr addr;
1047 
1048    cls = vai.ppc_icache_line_szB;
1049 
1050    /* Stay sane .. */
1051    vg_assert(cls == 16 || cls == 32 || cls == 64 || cls == 128);
1052 
1053    startaddr &= ~(cls - 1);
1054    for (addr = startaddr; addr < endaddr; addr += cls) {
1055       __asm__ __volatile__("dcbst 0,%0" : : "r" (addr));
1056    }
1057    __asm__ __volatile__("sync");
1058    for (addr = startaddr; addr < endaddr; addr += cls) {
1059       __asm__ __volatile__("icbi 0,%0" : : "r" (addr));
1060    }
1061    __asm__ __volatile__("sync; isync");
1062 
1063 #  elif defined(VGP_arm_linux)
1064    /* ARM cache flushes are privileged, so we must defer to the kernel. */
1065    Addr startaddr = (Addr) ptr;
1066    Addr endaddr   = startaddr + nbytes;
1067    VG_(do_syscall2)(__NR_ARM_cacheflush, startaddr, endaddr);
1068 
1069 #  elif defined(VGP_arm64_linux)
1070    // This arm64_linux section of this function VG_(invalidate_icache)
1071    // is copied from
1072    // https://github.com/armvixl/vixl/blob/master/src/a64/cpu-a64.cc
1073    // which has the following copyright notice:
1074    /*
1075    Copyright 2013, ARM Limited
1076    All rights reserved.
1077 
1078    Redistribution and use in source and binary forms, with or without
1079    modification, are permitted provided that the following conditions are met:
1080 
1081    * Redistributions of source code must retain the above copyright notice,
1082      this list of conditions and the following disclaimer.
1083    * Redistributions in binary form must reproduce the above copyright notice,
1084      this list of conditions and the following disclaimer in the documentation
1085      and/or other materials provided with the distribution.
1086    * Neither the name of ARM Limited nor the names of its contributors may be
1087      used to endorse or promote products derived from this software without
1088      specific prior written permission.
1089 
1090    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS CONTRIBUTORS "AS IS" AND
1091    ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
1092    WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
1093    DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE
1094    FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
1095    DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
1096    SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
1097    CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
1098    OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
1099    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
1100    */
1101 
1102    // Ask what the I and D line sizes are
1103    UInt cache_type_register;
1104    // Copy the content of the cache type register to a core register.
1105    __asm__ __volatile__ ("mrs %[ctr], ctr_el0" // NOLINT
1106                          : [ctr] "=r" (cache_type_register));
1107 
1108    const Int kDCacheLineSizeShift = 16;
1109    const Int kICacheLineSizeShift = 0;
1110    const UInt kDCacheLineSizeMask = 0xf << kDCacheLineSizeShift;
1111    const UInt kICacheLineSizeMask = 0xf << kICacheLineSizeShift;
1112 
1113    // The cache type register holds the size of the I and D caches as a power of
1114    // two.
1115    const UInt dcache_line_size_power_of_two =
1116        (cache_type_register & kDCacheLineSizeMask) >> kDCacheLineSizeShift;
1117    const UInt icache_line_size_power_of_two =
1118        (cache_type_register & kICacheLineSizeMask) >> kICacheLineSizeShift;
1119 
1120    const UInt dcache_line_size_ = 4 * (1 << dcache_line_size_power_of_two);
1121    const UInt icache_line_size_ = 4 * (1 << icache_line_size_power_of_two);
1122 
1123    Addr start = (Addr)ptr;
1124    // Sizes will be used to generate a mask big enough to cover a pointer.
1125    Addr dsize = (Addr)dcache_line_size_;
1126    Addr isize = (Addr)icache_line_size_;
1127 
1128    // Cache line sizes are always a power of 2.
1129    Addr dstart = start & ~(dsize - 1);
1130    Addr istart = start & ~(isize - 1);
1131    Addr end    = start + nbytes;
1132 
1133    __asm__ __volatile__ (
1134      // Clean every line of the D cache containing the target data.
1135      "0: \n\t"
1136      // dc : Data Cache maintenance
1137      // c : Clean
1138      // va : by (Virtual) Address
1139      // u : to the point of Unification
1140      // The point of unification for a processor is the point by which the
1141      // instruction and data caches are guaranteed to see the same copy of a
1142      // memory location. See ARM DDI 0406B page B2-12 for more information.
1143      "dc cvau, %[dline] \n\t"
1144      "add %[dline], %[dline], %[dsize] \n\t"
1145      "cmp %[dline], %[end] \n\t"
1146      "b.lt 0b \n\t"
1147      // Barrier to make sure the effect of the code above is visible to the rest
1148      // of the world.
1149      // dsb : Data Synchronisation Barrier
1150      // ish : Inner SHareable domain
1151      // The point of unification for an Inner Shareable shareability domain is
1152      // the point by which the instruction and data caches of all the processors
1153      // in that Inner Shareable shareability domain are guaranteed to see the
1154      // same copy of a memory location. See ARM DDI 0406B page B2-12 for more
1155      // information.
1156      "dsb ish \n\t"
1157      // Invalidate every line of the I cache containing the target data.
1158      "1: \n\t"
1159      // ic : instruction cache maintenance
1160      // i : invalidate
1161      // va : by address
1162      // u : to the point of unification
1163      "ic ivau, %[iline] \n\t"
1164      "add %[iline], %[iline], %[isize] \n\t"
1165      "cmp %[iline], %[end] \n\t"
1166      "b.lt 1b \n\t"
1167      // Barrier to make sure the effect of the code above is visible to the rest
1168      // of the world.
1169      "dsb ish \n\t"
1170      // Barrier to ensure any prefetching which happened before this code is
1171      // discarded.
1172      // isb : Instruction Synchronisation Barrier
1173      "isb \n\t"
1174      : [dline] "+r" (dstart),
1175        [iline] "+r" (istart)
1176      : [dsize] "r" (dsize),
1177        [isize] "r" (isize),
1178        [end] "r" (end)
1179      // This code does not write to memory but without the dependency gcc might
1180      // move this code before the code is generated.
1181      : "cc", "memory"
1182    );
1183 
1184 #  elif defined(VGA_mips32) || defined(VGA_mips64)
1185    SysRes sres = VG_(do_syscall3)(__NR_cacheflush, (UWord) ptr,
1186                                  (UWord) nbytes, (UWord) 3);
1187    vg_assert( !sr_isError(sres) );
1188 
1189 #  endif
1190 }
1191 
1192 
1193 /* ---------------------------------------------------------------------
1194    dcache flushing
1195    ------------------------------------------------------------------ */
1196 
VG_(flush_dcache)1197 void VG_(flush_dcache) ( void *ptr, SizeT nbytes )
1198 {
1199    /* Currently this is only required on ARM64. */
1200 #  if defined(VGA_arm64)
1201    Addr startaddr = (Addr) ptr;
1202    Addr endaddr   = startaddr + nbytes;
1203    Addr cls;
1204    Addr addr;
1205 
1206    ULong ctr_el0;
1207    __asm__ __volatile__ ("mrs %0, ctr_el0" : "=r"(ctr_el0));
1208    cls = 4 * (1ULL << (0xF & (ctr_el0 >> 16)));
1209 
1210    /* Stay sane .. */
1211    vg_assert(cls == 64 || cls == 128);
1212 
1213    startaddr &= ~(cls - 1);
1214    for (addr = startaddr; addr < endaddr; addr += cls) {
1215       __asm__ __volatile__("dc cvau, %0" : : "r" (addr));
1216    }
1217    __asm__ __volatile__("dsb ish");
1218 #  endif
1219 }
1220 
1221 /*--------------------------------------------------------------------*/
1222 /*--- end                                                          ---*/
1223 /*--------------------------------------------------------------------*/
1224