thread safe path comparison or canonicalization function for Unix?

Question

I'd seen some ancient code that simplifed Unix paths for comparison by doing something like the following pseudocode:

• strip off the last path component to get a directory only part: /foo/bar -> /foo
• getcwd and remember original path
• chdir /foo
• getcwd and remember to return to caller
• chdir old original path

Is there a standard Unix system function that does all this without the thread unsafe current directory manipulation?

A mutex could make that code sequence less thread unsafe, but isn't really ideal (you'd have to know that all other code using getcwd or other functions dependent on the process cwd including system and vendor code protects with this same mutex).

Answer 1

Oh dear, doing the action you mention couldn't possibly be thread safe, because it actually chdir's, which is going to confuse any other threads. I'll have to look up the string-manipulation portion of what you want, but it can't possibly also strip softlinks or do anything else that requires asking the operating system for file information without being a little thread-unsafe.

Related posts:

• http://stackoverflow.com/questions/147057/unix-path-searching-c-function
• http://stackoverflow.com/questions/2128775/how-to-parse-a-folder-path-with-spaces-in-c-code

Try this to convert relative file paths, then compare them as strings:

#include<stdio.h>
#include<dirent.h>
#include<fcntl.h>
#include<sys/param.h>
int main( int argc, char **argv )
{
  char buffer[MAXPATHLEN+1];
  if( argc <= 1 ) return 0;
  DIR*d = opendir( argv[1] );
  if( !d ) return 0;
  int dfd = dirfd(d);
  if( !dfd ) return 0;
  int result = fcntl( dfd, F_GETPATH, buffer );
  if( result == -1 ) return 0;
  fprintf( stdout, "path='%s'\n", buffer );
  return 0;
}

Answer 2

Try realpath() or canonicalize_file_name()

If your system supports it (and it probably does), I suggest calling realpath(pathname, NULL); this will malloc the buffer for the canonicalized filename and pass it back as the return value. You'd have to be sure to free() the pointer. The alternative, passing in an output buffer, runs the risk of buffer overruns.

canonicalize_file_name() is a Gnu extension that is equivalent to realpath(pathname, NULL).

Answer 3

There is no "canonical" path in a Unix directory. It may be possible for a file/directory to have multiple hard links/mount points.

The closest thing to the identity of a file/directory is its inode.

Answer 4

What about realpath(3)?

Since it returns its result in a buffer you supply, thread-safety should not be an issue.