Andrew Morgan wrote:
>
> Chris Jaeger wrote:
> > Would a patch substituting getpwnam_r for getpwnam be accepted?
>
> I think I'd like to see an autoconf test for this, and something to
> ensure that any module that uses it can fall back to the non _r function
> if its not available at compile time.
>
While I might write such a patch for my own use, I'm not
sure submitting it will be worthwhile. The half-way fix is simply
to separate pam_unix's struct passwd buffer from the one used by
libc's getpwnam. Sufficient for my needs, but it doesn't solve any
fundamental problems. Making pam_unix reentrant would be more
useful in general, but that would require a bit more work. Or maybe
I'll just fall back on pam_pwdb, which is also sufficient. If
anyone wants the half-way "fix", let me know.
Just mumbling to myself... (in case someone cares to point out
something that I've missed/something idiotic I've said)
Thanks,
Chris
Things that would need to be done to make pam_unix
completely reentrant (items may be missing):
unix_chkpwd.c: getuidname returns a static local pointer.
support.c: PAM_getlogin contains a static local
pam_unix_passwd.c: contains several static local vars
bigcrypt.c: static storage area (but this appears to be by
mandate)
*.c: several calls to getpwnam
