Using today's rawhide (20080421, kernel-2.6.25-1.fc9.i586) I get a kernel BUG when the system attempts to mount its nfs install source. anaconda says: Running... /bin/mount -n -t nfs -o nolock [nfs target] /mnt/isodir And then we get: kernel BUG at security/selinux/hooks.c:761! ... EIP is at selinux_sb_clone_mnt_opts+0x21/0xd9 ... Call Trace: [I'm trimming it to just function names] security_sb_clone_mnt_opts nfs_xdev_get_sb vfs_kern_mount nfs_follow_mountpoint inode_has_perm selinux_inode_follow_link __link_path_walk inode_has_perm avc_has_perm path_walk do_path_lookup __path_lookup_intent_open path_lookup_open open_namei __dentry_open do_filp_open strncpy_from_user do_sys_open fput sys_open syscall_call ht_enable_msi_mapping This causes loader to SEGV and the install dies.
Well, I guess if the installer just isn't loading policy or disabling selinux then we could hit this BUG_ON(!ss_initialized) sds, what do you think, if !ss_initialized we know the parent sb couldn't have had selinux mount options so there is no reason we can't just throw this sb on the superblock_security_head list and deal with it later... will send such a patch to selinux list in just a moment if I don't hear anything different.
Created attachment 303195 [details] change BUG_ON to safe handling if (!ss_initialized)
Looks sane, although I'm curious as to why we haven't loaded a policy before we hit the clone.
policy is part of the second stage, and isn't loaded until after it's mounted.
I'm going to push the patch in comment #2 to F-9 kernel as soon as CVS opens back up
Patch went in kernel 2.6.25-4. Can someone confirm this is fixed?
I can't easily confirm this until a fixed kernel lands in rawhide - it needs to be running during the install. I'll see if it's possible to build custom boot images using the koji repos..
Well, I built custom boot images with the new kernel and attempted an NFS install. It didn't oops on NFS mount, and it seemed to mount stage2.img OK as well. Unfortunately the tree is out-of-date so the install can't proceed past there. But I think that confirms that the BUG is gone.