[sanitizer_common] Use O_TRUNC for WrOnly access mode.

[MaskRay]

Summary

Otherwise if the file existed and was larger than the write size before the OpenFile call, the file will not be truncated and contain garbage in trailing bytes.

[vitalybuka]

Could you please add corresponding check into "TEST(SanitizerCommon, FileOps)" and make sure that all platforms have the same behavior.

[MaskRay]

Add test

[MaskRay]

Added test. BTW, is my testing correct?

% ninja -C ~/Dev/llvm/release TSanitizer-i386-Test TSanitizer-x86_64-Test TSanitizer-x86_64-Test-Nolibc && ~/Dev/llvm/release/projects/
compiler-rt/lib/sanitizer_common/tests/Sanitizer-i386-Test && ~/Dev/llvm/release/projects/compiler-rt/lib/sanitizer_common/tests/Sanitiz
er-x86_64-Test
# passed

[vitalybuka]

In D48250#1134404, @MaskRay wrote:

Added test. BTW, is my testing correct?

% ninja -C ~/Dev/llvm/release TSanitizer-i386-Test TSanitizer-x86_64-Test TSanitizer-x86_64-Test-Nolibc && ~/Dev/llvm/release/projects/
compiler-rt/lib/sanitizer_common/tests/Sanitizer-i386-Test && ~/Dev/llvm/release/projects/compiler-rt/lib/sanitizer_common/tests/Sanitiz
er-x86_64-Test
# passed

following should run the test

ninja -C ~/Dev/llvm/release check-sanitizer

[vitalybuka]

lib/sanitizer_common/sanitizer_posix.cc

163

please also fix
compiler-rt/lib/sanitizer_common/sanitizer_rtems.cc

windows looks good
compiler-rt/lib/sanitizer_common/sanitizer_win.cc

[MaskRay]

Add other platforms as vitalybuka suggested

[MaskRay]

Thanks for informing me of the check-sanitizer target. 64 SanitizerCommon-lsan-i386-Linux tests already fail before this revision. All other tests pass.

[vitalybuka]

lib/sanitizer_common/sanitizer_solaris.cc

81 ↗

(On Diff #151600)

This function has different signature.

@krytarowski I don't see any calls to this function. Is this here accidentally?

[MaskRay]

Remove change to solaris which seems already broken.

[MaskRay]

lib/sanitizer_common/sanitizer_solaris.cc

81 ↗

(On Diff #151600)

I don't know much about the file history. But I guess the signature of OpenFile has changed from two parameters to three parameters at some point and the solaris version does not catch up.

// Returns kInvalidFd on error.
fd_t OpenFile(const char *filename, FileAccessMode mode,
              error_t *errno_p = nullptr);

Just reverted the change to the solaris file.

[krytarowski]

lib/sanitizer_common/sanitizer_solaris.cc

81 ↗

(On Diff #151600)

Hmm, CC @ro. I'm not sure about SunOS.

lib/sanitizer_common/sanitizer_posix.cc

 #if !SANITIZER_MAC
 void MprotectMallocZones(void *addr, int prot) {}
 #endif
 
 fd_t OpenFile(const char *filename, FileAccessMode mode, error_t *errno_p) {
   int flags;
   switch (mode) {
     case RdOnly: flags = O_RDONLY; break;
-    case WrOnly: flags = O_WRONLY | O_CREAT; break;
+    case WrOnly: flags = O_WRONLY | O_CREAT | O_TRUNC; break;
     case RdWr: flags = O_RDWR | O_CREAT; break;
   }
   fd_t res = internal_open(filename, flags, 0660);
   if (internal_iserror(res, errno_p))
     return kInvalidFd;
   return res;
 }
 

lib/sanitizer_common/tests/sanitizer_libc_test.cc

TEST(SanitizerCommon, FileOps) {
   uptr len1 = internal_strlen(str1);
   const char *str2 = "zxcv";
   uptr len2 = internal_strlen(str2);
 
   char tmpfile[128];
   temp_file_name(tmpfile, sizeof(tmpfile), "sanitizer_common.fileops.tmp.");
   fd_t fd = OpenFile(tmpfile, WrOnly);
   ASSERT_NE(fd, kInvalidFd);
+  ASSERT_FALSE(internal_iserror(internal_write(fd, "A", 1)));
+  CloseFile(fd);
+
+  fd = OpenFile(tmpfile, WrOnly);
+  ASSERT_NE(fd, kInvalidFd);
+  EXPECT_EQ(internal_lseek(fd, 0, SEEK_END), 0u);
   uptr bytes_written = 0;
   EXPECT_TRUE(WriteToFile(fd, str1, len1, &bytes_written));
   EXPECT_EQ(len1, bytes_written);
   EXPECT_TRUE(WriteToFile(fd, str2, len2, &bytes_written));
   EXPECT_EQ(len2, bytes_written);
   CloseFile(fd);
 
   EXPECT_TRUE(FileExists(tmpfile));