[PATCH] [sanitizer] Fix log_path behavior with StopTheWorld.

[Sergey Matveev]

  > Do they have the shared memory (and thus shared report_fd/report_fd_pid, but different internal_getpid())?

  Yes.

  They don't have to simultaneously call RawWrite for it to be a problem. Consider this possibility: the parent opens a log file, writes some data to it. Then the child suspends the parent, opens its own log file (closing the parent's file). The child then exits and the parent resumes execution. If the parent now attempts to write more log messages, it will reopen its old log file and overwrite whatever was written to it earlier.

http://llvm-reviews.chandlerc.com/D2306