On Wed, 12 Sep 2001, Thorsten Kukuk wrote:
> > Is the following ok for a first implementation, or do you have more ambitious
> > plans? I believe you mentioned providing a wrapper for systems that don't
> > have getpwnam_r(), but I'm personally quite content with this.
> You should not allocate a static buffer for use with getpwnam_r.
> Instead query the return code and if errno = ERANGE (I think this was
> the value) increase the buffer and try it again.
> Else the result will be worse then the current situation, longer entries
> will fail.
Gotcha. Take two, then. Certainly clumsier than a single call to getpwnam(),
but hopefully this gets us started in the right direction.
I've committed this particular change already on the head CVS branch, in
modules/pam_unix/support.c -- which coincidentally doesn't include
_pam_aconf.h, and therefore doesn't see the HAVE_GETPWNAM_R define yet.
Further corrections are welcome.
Steve Langasek
postmodern programmer
struct passwd *pwd = NULL;
#if HAVE_GETPWNAM_R
char *buf = NULL;
int bufsize = 0;
struct passwd pwd_buf;
pwd = &pwd_buf;
#endif
...
#if HAVE_GETPWNAM_R
bufsize = 1024;
buf = malloc(bufsize);
if ((retval = getpwnam_r(name, pwd, buf, bufsize, &pwd))) {
pwd = NULL;
}
while (retval == ERANGE) {
bufsize += 1024;
buf = realloc(buf, bufsize);
if ((retval = getpwnam_r(name, pwd, buf, bufsize, &pwd))) {
pwd = NULL;
}
}
#else
pwd = getpwnam(name);
#endif
...
#if HAVE_GETPWNAM_R
if (buf)
free (buf);
#endif
return;
