/etc/init.d/halt.sh includes a step to remount all filesystems read only before shutting down or rebooting. On kernel 2.4.25-gentoo and all earlier kernels that I have used (2.4.22-gentoo and various others) this worked fine. Upgrading to 2.4.26-gentoo-r3 (and r5) this step fails. the remount causes / to be inaccessable so the script cannot access what it needs to finish the shutdown. While this is a kernel bug I'll attach a patch to halt.sh that works around it. Reproducible: Always Steps to Reproduce: 1. use the stated kernel with nfsroot (maybe any nfs, i haven't tried) 2. remount nfsroot / as ro 3. enjoy your hosed system
Created attachment 35523 [details, diff] workaround patch for halt.sh
Could you please try vanilla-sources-2.4.26 to see if we can isolate this to a patch in the gentoo-sources patchset? Thanks...
Okay, i've tried a few kernels now for the halt.sh nfsroot remount,ro problem: 2.4.26-vanilla: does NOT have the problem 2.4.26-gentoo-r3 & r5: has the problem 2.6.7-gentoo-r11: has the problem i'll attach my .config files for each kernel
Created attachment 35619 [details] 2.4.26-gentoo-r5 .config that exhibits the problem
Created attachment 35620 [details] 2.6.7-gentoo-r11 .config that exhibits the problem
Created attachment 35621 [details] vanilla 2.4.26 .config that does NOT exhibit the problem
Can you try a similar 2.4 .config on both kernels - there are quite a few =Y <> =M changes between the two which might be causing somehow. But since you're getting this on gentoo-dev-sources too; I suspect this might be due to supermount. Can you please try: UNIPATCH_EXCLUDE="02-02.superMount.1.2.11a.patch" emerge gentoo-sources and see how that goes? Thanks...
Greg, are you able to do the test requested by Tim? Since it's really a kernel problem I'm really hesitant to apply any patches to halt.sh to workaround
sorry, i haven't had a chance to build a new kernel yet. i suspect its a supermount bug/issue as that makes the most sense. i'll follow up once i build a new kernel w/o supermount.
Bump... please try a new and clean kernel (no supermount), and reopen this bug if the issue still exists.