perfMemory_bsd.cpp get_user_name_slow misses check ?

[gerard ziemski]

hi Matthias,

Looking at "open_directory_secure()" API, which is called just before 
"is_directory_secure()", seems to suggest that it is the 
"is_directory_secure()" call that is not needed here.

So perhaps linux/win/solaris have it wrong and bsd got it right?


// Open the directory of the given path and validate it.
// Return a DIR * of the open directory.
//
static DIR *open_directory_secure(const char* dirname) {
   // Open the directory using open() so that it can be verified
   // to be secure by calling is_dirfd_secure(), opendir() and then check
   // to see if they are the same file system object.  This method does not
   // introduce a window of opportunity for the directory to be attacked 
that
   // calling opendir() and is_directory_secure() does.



cheers




On 2/3/20 8:02 AM, Baesken, Matthias wrote:
> Hello,  I noticed that  perfMemory_bsd.cpp function
>
> static char* get_user_name_slow(int vmid, TRAPS)
>
> misses  the
>
>      // Since we don't create the backing store files in directories
>      // pointed to by symbolic links, we also don't follow them when
>      // looking for the files. We check for a symbolic link after the
>      // call to opendir in order to eliminate a small window where the
>      // symlink can be exploited.
>      //
>      if (!is_directory_secure(usrdir_name)) {
>        FREE_C_HEAP_ARRAY(char, usrdir_name);
>        os::closedir(subdirp);
>        continue;
>      }
>
>
> Check that  aix/linux/solaris    have .
> Should this be added ?
>
>
> Best Regards, Matthias