Fix getting environment variables for sanitizers needs on FreeBSD

[emaste]

lib/sanitizer_common/sanitizer_linux.cc

325 ↗

(On Diff #10693)

Note that ::environ may be null - see the libc implementation here:
http://svnweb.freebsd.org/base/head/lib/libc/stdlib/getenv.c?revision=253413&view=markup#l428

[samsonov]

lib/sanitizer_common/sanitizer_linux.cc

325 ↗

(On Diff #10693)

Why can't you use internal_strstr function here?

[kutuzov.viktor.84]

lib/sanitizer_common/sanitizer_linux.cc

325 ↗

(On Diff #10693)

I believe strstr() doesn't look up through an array of pointers? Or, if you mean to replace internal_strncmp(*Env, name, NameLen) == 0 with internal_strstr(*Env, name) == *Env, well, it would look a bit awkward to me. But no problem to fix, of course.

[samsonov]

lib/sanitizer_common/sanitizer_linux.cc

325 ↗

(On Diff #10693)

Ah, sorry, please disregard my comment.

[kutuzov.viktor.84]

Dmitry, your commit rL211417 does the same, but makes things libc-dependent. Do you think we should drop down this patch or replace your implementation? Thanks.

Update: getenv() relies on strncmp() and other intercepted funcs and thus breaks Asan initialization.

[dvyukov]

I all in favor of removing libc dependencies.
But where does "extern char **environ" came from? Doesn't it come from libc?

[emaste]

But where does "extern char **environ" came from? Doesn't it come from libc?

char **environ is provided by crt1 - it's here on FreeBSD -HEAD:
http://svnweb.freebsd.org/base/head/lib/csu/common/ignore_init.c?revision=245133&view=markup

[dvyukov]

I see.
Does the current version work on freebsd?
What's the purpose of removing libc dependency from there?

crt1 is a part of C runtime. I am interested in complete independence of C world for Go runtime.
But I see that now it's dependent on lib anyway, e.g. there is a call to pthread_self.

[kutuzov.viktor.84]

Does the current version work on freebsd?

Yes, though I need to update the patch to handle the case with null 'environ' first. Will do ASAP.

What's the purpose of removing libc dependency from there?

AFAIU, the less we tied to libc in things like that, the less the need to handle under-initialized sanitizers in intercepted functions.

crt1 is a part of C runtime. I am interested in complete independence of C world for Go runtime.

We would need a special implementation for Go then, I believe.

[kutuzov.viktor.84]

Updated to reflect recent changes and to test ::environ for NULL.

[dvyukov]

lib/sanitizer_common/sanitizer_linux.cc

354 ↗

(On Diff #11145)

s/NULL/0/ here in elsewhere
we don't use NULL in sanitizer runtimes, it may not even be defined here

356 ↗

(On Diff #11145)

this won't work w/o include <stdlib.h> that you removed
replace it with #error "unsupported platform"

[kutuzov.viktor.84]

Updated as suggested.

[kutuzov.viktor.84]

Closed by commit rL212690 (authored by vkutuzov).

compiler-rt/trunk/lib/sanitizer_common/sanitizer_linux.cc

 #include <sys/resource.h>
 #include <sys/stat.h>
 #include <sys/syscall.h>
 #include <sys/time.h>
 #include <sys/types.h>
 #include <unistd.h>
 
 #if SANITIZER_FREEBSD
-#include <stdlib.h>  // for getenv
 #include <sys/sysctl.h>
 #include <machine/atomic.h>
 extern "C" {
 // <sys/umtx.h> must be included after <errno.h> and <sys/types.h> on
 // FreeBSD 9.2 and 10.0.
 #include <sys/umtx.h>
 }
+extern char **environ;  // provided by crt1
 #endif  // SANITIZER_FREEBSD
 
 #if !SANITIZER_ANDROID
 #include <sys/signal.h>
 #endif
 
 #if SANITIZER_ANDROID
 #include <android/log.h>

 #else
   kernel_timeval tv;
 #endif
   internal_memset(&tv, 0, sizeof(tv));
   internal_syscall(SYSCALL(gettimeofday), (uptr)&tv, 0);
   return (u64)tv.tv_sec * 1000*1000*1000 + tv.tv_usec * 1000;
 }
 
-#if SANITIZER_LINUX
-// Like getenv, but reads env directly from /proc and does not use libc.
-// This function should be called first inside __asan_init.
+// Like getenv, but reads env directly from /proc (on Linux) or parses the
+// 'environ' array (on FreeBSD) and does not use libc. This function should be
+// called first inside __asan_init.
 const char *GetEnv(const char *name) {
+#if SANITIZER_FREEBSD
+  if (::environ != 0) {
+    uptr NameLen = internal_strlen(name);
+    for (char **Env = ::environ; *Env != 0; Env++) {
+      if (internal_strncmp(*Env, name, NameLen) == 0 && (*Env)[NameLen] == '=')
+        return (*Env) + NameLen + 1;
+    }
+  }
+  return 0;  // Not found.
+#elif SANITIZER_LINUX
   static char *environ;
   static uptr len;
   static bool inited;
   if (!inited) {
     inited = true;
     uptr environ_size;
     len = ReadFileToBuffer("/proc/self/environ",
                            &environ, &environ_size, 1 << 26);
   }
   if (!environ || len == 0) return 0;
   uptr namelen = internal_strlen(name);
   const char *p = environ;
   while (*p != '\0') {  // will happen at the \0\0 that terminates the buffer
     // proc file has the format NAME=value\0NAME=value\0NAME=value\0...
     const char* endp =
         (char*)internal_memchr(p, '\0', len - (p - environ));
     if (endp == 0)  // this entry isn't NUL terminated
       return 0;
     else if (!internal_memcmp(p, name, namelen) && p[namelen] == '=')  // Match.
       return p + namelen + 1;  // point after =
     p = endp + 1;
   }
   return 0;  // Not found.
-}
 #else
-const char *GetEnv(const char *name) {
-  return getenv(name);
-}
+#error "Unsupported platform"
 #endif
+}
 
 extern "C" {
   SANITIZER_WEAK_ATTRIBUTE extern void *__libc_stack_end;
 }
 
 #if !SANITIZER_GO
 static void ReadNullSepFileToArray(const char *path, char ***arr,
                                    int arr_size) {