Jeremy Huntwork wrote:
Alexander E. Patrakov wrote:
If all three points are true, this proves with 100% fidelity that the
bug is not in the init, and is caused by early /sbin/hotplug processes.
Unionfs bug, in other words. Let's act accordingly and admit that my
idea with unionfs was wrong. Possible solutions:
Certainly this could be the problem. But how do we know absolutely that
the problem is indeed unionfs? I haven't seen that explained. Before we
make another decision concerning this, I'd like to see information that
shows without a shadow of doubt this is a unionfs bug.
Let's do a direct experiment: replace unionfs with a bind mount. This
will, obviously, produce a non-functional (read-only) CD, where even
udev bootscript will fail, but if unionfs was indeed the problem, init
will at least print its version.
Justin: could you please take the 6.1.1-3 CD and rebuild only the init,
with the attached patch? The kernel and the .root.sqfs file should stay
the same as on the original CD.
If this prints the init version on a computer where the original 6.1.1-3
CD locks up, this would show without any doubt that the lockup is in
unionfs.
BTW, since we have reports about the lockup on 6.2-pre3, and because (as
already explained in
http://linuxfromscratch.org/pipermail/livecd/2006-February/003067.html)
the changes were not well-reasoned, I think that r1387 should be reverted.
--
Alexander E. Patrakov
Index: initramfs/init.c
===================================================================
--- initramfs/init.c (revision 1411)
+++ initramfs/init.c (working copy)
@@ -119,7 +119,7 @@
/* Activate unionfs */
printf("Mounting unionfs...\n");
- i = mount("unionfs", "/.union", "unionfs", 0, "dirs=/.tmpfs/.overlay=rw:/.tmpfs/.cdrom=ro:/.tmpfs/.sqfs=ro");
+ i = mount(SQFS, "/.union", NULL, MS_BIND, NULL);
if (i<0) {
printf("Failed to mount unionfs: %s\n", strerror(errno));
return(0);
--
http://linuxfromscratch.org/mailman/listinfo/livecd
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page
