A question on the implementation of /dev/stdin, etc.

[Kevin Schoedel]

This discussion has prompted me to try to implement /dev/fd/... for
System V 3.2 on a 386.  I don't have kernel source, so I can't use the
   case IFCHR:
      if( major( dev ) == fd_dev ){
          dup minor(dev)
          return;
      }
      (*cdevsw[ ....
technique that was mentioned here.  I don't want to use a file system
for /dev/fd/... because that would prevent me using the synonyms
/dev/stdin, etc.  So... is there any fundamental problem with having
a device, consisting only of an open routine, that gets rid of the
file descriptor just created for /dev/fd/N and then dup's N?
Specifically, I tried an open function containing, roughly,
   f = ...file descriptor for /dev/fd/...
   close( f )
   dup( minor( dev ) )
As I said, I don't have source, and my knowledge of the kernel implementation
is restricted to having read John Lions' book, and perhaps things have
changed a bit since version 6.  First, I tried looking in u.u_ar0[0] for
the file descriptor, but that was zero, which didn't seem right.
Then I tried searching u.u_ofile[] for a file whose inode was that of the
special file -- and sure enough, it was in u.u_ofile[0].  So now I'm
confused.  Am I doing something slightly wrong or fundamentally wrong?

-kevin