Re: libio/libioP.h:ALLOC_BUF() unconditionally uses mmap()

[Wayne Whitney]

P.S.  On the separate topic of modifying MAGMA instead of glibc or the
kernel, it occured to me that since MAGMA is linked statically, it could
just redefine __mmap() to request a higher address.  So I tried adding
this to a test program:

  #define __NR_real_mmap __NR_mmap
  #define NEW_MAP_FIXED_BASE 0xA0000000

  static inline _syscall6(void *, real_mmap, void *, start, size_t, len, int, 
prot, int, flags, int, fd, off_t, offset);

  void *__mmap(void *start, size_t len, int prot, int flags, int fd, off_t 
offset)
  {

      if (start == 0 && fd == -1) {
          start = (void *) NEW_MAP_FIXED_BASE;
          /* flags &= MAP_FIXED; Is this needed */
      }
      return (real_mmap(start, len, prot, flags, fd, offset));
  }

That changed the output of strace on the program from:

  old_mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) 
= 0x40000000

to:

  old_mmap(ptrace: umoven: Input/output error
  ptrace: umoven: Input/output error
  )                              = -1 EFAULT (Bad address)

which suggests that I did something wrong.  :-)  I guess the mmap syscall
does not follow the syscall calling style that the _syscall6 macro
assumes.  Is there an eaiser way to do this than to duplicate the contents
of glibc-2.2.4/sysdeps/unix/sysv/linux/i386/mmap.S?  I assume that once I
have defined __mmap in my program, there is no way to reference the __mmap
object from mmap.o in libc.a.

Thanks, Wayne