Hi,
(The "use-case" here being that for some funky reason dbus needs to get
started in order to build core parts of Gnome in Debian, thus Gnome
is broken in Debian GNU/Hurd until this is fixed, or we decide to build
Gnome completely manually from now on)
The following program works fine in a real installation, but fails with
EADDRNOTAVAIL if run in a chroot:
#include <sys/socket.h>
#include <sys/un.h>
#include <errno.h>
#include <string.h>
#include <stdio.h>
#define SOCKET_ADDR "/tmp/foo"
int main(int argc, char** argv)
{
int fd;
struct sockaddr_un addr;
size_t path_len;
addr.sun_family = AF_UNIX;
path_len = strlen(SOCKET_ADDR);
strncpy(addr.sun_path, SOCKET_ADDR, path_len+1);
fd = socket(PF_UNIX, SOCK_STREAM, 0);
if (bind(fd, (struct sockaddr*) &addr, SUN_LEN(&addr)) < 0)
printf("Error: %s\n", strerror(errno));
}
(This code is a minimal reduction of the dbus-daemon startup code, which
fails in the same way)
The problem is apparently that /hurd/ifsock is set as a passive
translator in [glibc]/sysdeps/mach/hurd/bind.c via
__file_set_translator() and is not aware that it is started in a chroot:
<marcus> azeem: it seems that ifsock talks to a different pflocal than
the application
<marcus> azeem: could it be that it escapes the chroot somehow?
<azeem> what is "it"? ifsocks? the application?
<marcus> yes
<marcus> ifsocks
<marcus> one or the other ;)
<youpi> the starting of the translator ? (that possibly doesn't even
know that there is chroot at stake)
<marcus> ah, hm
<marcus> youpi: yeah
<youpi> how should it know that anyway ?
<marcus> well
<marcus> youpi: oh
<marcus> you mean libc sets a passive translator?
<marcus> right
Otherwise, maybe this is a problem with the chroot setup, marcus
suggested setting a firmlink to the main pflocal as a work around, but I
was not successful in doing that (ordinary pipes work in the chroot,
by the way)
Is this an error in glibc or by myself? Can anybody suggest a
work-around maybe, so that Gnome can be built again?
Michael
_______________________________________________
Bug-hurd mailing list
Bug-hurd@gnu.org
http://lists.gnu.org/mailman/listinfo/bug-hurd
