Le vendredi 02 décembre 2005 à 14:38 -0500, Stephen Smalley a écrit :
On Fri, 2005-12-02 at 20:20 +0100, Nicolas Mailhot wrote:
Le vendredi 02 décembre 2005 à 14:17 -0500, Stephen Smalley a écrit :
It isn't the number of nodes in /dev; it is the number of entries in file_contexts. And the slowdown should be improved/eliminated with recent changes in libselinux (1.27.28); let us know if it isn't. There are two changes in libselinux, one of which will have immediate benefit without requiring any changes to udev, and the other of which requires a small change to udev to take advantage of.
BTW today's rawhide segfaults on boot if run in enforcing mode
checkpolicy-1.27.19-1 selinux-policy-targeted-2.0.7-2 audit-1.1.1-1 audit-libs-1.1.1-1 audit-libs-1.1.1-1 libselinux-1.27.28-1 libselinux-1.27.28-1 libsepol-1.9.41-1 libsepol-1.9.41-1 libsemanage-1.3.61-1
Adding selinux=false to the boot arguments rescues the system
Hmmm...same versions of the above, don't see this behavior. Details?
No :( Did a rawhide update (kernel + selinux stuff), touch ./autorelabel, reboot -> bang Tried the previous working kernel -> bang Rebooted on the rescue disk, nothing in the system logs (crash too early at selinux init)
Since the last sync was only selinux-related, decided to try selinux=false before mucking with the system, and everything booted at once.
Will try to reproduce now, in case it was a transient problem