Stephen Smalley wrote:
On Fri, 2004-09-03 at 11:43, Tom London wrote:
>Newest Rawhide packages improve things a bit for strict/enforcing, but
>still no joy.
>
>When booting strict/enforcing, the system seems to boot to single user mode,
>but is unable to write to the console. Last messages are avc denials from
>/bin/dmesg, that seem to occur just before the 'Welcome to Fedora' message.
>I can hear the device discovery going on, but nothing on the console.
>After about 5 minutes, ALT-CTL-DEL brought the system down, with the
>customary console messages. (But, error messages about most file systems
>not being mounted).
>
>Here are the early avcs...
>
>Sep 3 07:25:35 fedora kernel: audit(1094196259.050:0): avc: denied {
>create } for pid=1 exe=/sbin/init name=initctl
>scontext=system_u:system_r:init_t tcontext=system_u:object_r:unlabeled_t
>tclass=fifo_file
>Sep 3 07:25:36 fedora smartd[2856]: Opened configuration file
>/etc/smartd.conf
>Sep 3 07:25:36 fedora kernel: audit(1094196259.050:0): avc: denied {
>associate } for pid=1 exe=/sbin/init name=initctl
>scontext=system_u:object_r:unlabeled_t tcontext=system_u:object_r:fs_t
>tclass=filesystem
>
>
No point in even trying to work from those audit messages, as the tmpfs
entry in fs_use in the rawhide policy is wrong and will break all users
of anonymous shared mappings and System V shared memory regardless of
whether it ever works for tmpfs /dev.
And life is still rather unpleasant even if fs_use is reverted to the
upstream policy. Using fscontext=system_u:object_r:device_t on the
tmpfs /dev mount would help significantly, but the claim is that it is
mounted before the initial policy load. End result is that tmpfs_t ends
up doing double duty as a type on shmem and /dev, which has a huge
impact on existing policy.
Strongly advise changing initialization to umount the initial tmpfs /dev
prior to initrd exit and re-mount it _after_ the initial policy load
using fscontext=. Or load a minimal policy from the initrd in your
/linuxrc prior to original tmpfs mount.
Most of the problems with booting strict SELinux with /dev/ mounted on a
tmpfs file system should be fixed by the
latest policy and initscripts package.
Dan