On Thu, Aug 29, 2002 at 10:35:31AM +0300, Giorgos Keramidas wrote:
+> Hello,
+>
+> A friend asked me why he was getting warnings about conversion of
+> unsigned to signed, when calling chown() with:
+>
+> chown("/dev/null", -1, -1);
+>
+> The manpage of chown has a prototype of:
+>
+> int
+> chown(const char *path, uid_t owner, gid_t group);
+>
+> But the manpage mentions that (uid_d)-1 is the right value for a chown
+> operation that wants to leave the user part of the owner:group set
+> unchanged.
+>
+> After a bit of research he found that the definition of uid_t was
+> (unsigned int) and that's why -1 was giving him warnings. But the fun
+> doesn't stop there... The implementation of chown in vfs_syscalls.c
+> uses `int' as the type of the two uid_t/gid_t arguments in the *uap
+> argument, and then calls setfown() which accepts uid_t and gid_t !!!
+>
+> Then setfown() copies the values in a `struct vattr' at the va_uid and
+> va_gid members, which are also uid_t and gid_t.
+>
+> The only place where the uid & gid arguments of chown() are int and
+> not uid_t and gid_t is in the prototype of the system call itself.
+> Is this really necessary? Is there a reason behind it?
Yes, look at VOP_SETATTR man page:
if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
/*
* Change owner and/or group of the file.
*/
}
And:
leila:root:~# egrep "#define.*VNOVAL" /sys/sys/vnode.h
/sys/sys/vnode.h:#define VNOVAL (-1)
--
Pawel Jakub Dawidek
UNIX Systems Administrator
http://garage.freebsd.pl
Am I Evil? Yes, I Am.
msg36579/pgp00000.pgp
Description: PGP signature
