On Wednesday 05 April 2017 03:14 PM, Xunlei Pang wrote:
Why do we have to introduce a new "force_no_rebuild"? As
far as I know it's mainly for
diskless machines, how about we detect this case(a. the initramfs directory is RO, b.
there
is one valid kdump initramfs), then we always do the "kexec -p" reload process
and throw
some message to notify users about it?
Current implementation is that if /boot is RO and there exists a rebuild
condition,we generate a warning and FAIL.
There is a risk if we change that in warning + PASS.
As Dave has already said about it:
"There could be risk to silently go ahead for a readonly mounted /boot,
sometimes a corrupt disk will be mounted as ro automatically or for some
other reason one do not intend to do this. For those cases failout
should be better."
Or at least I think we can do it in case of RO nfs mount, or firstly check
/etc/fstab(/boot is
often defined in it) options, if it is RO in /etc/fstab, we can safely do the "kexec
-p" reload.
I am not sure if diskless system can always have /etc/fstab, but
anyway...there can be some other way to know if /boot is RO. Actually,
this force_no_rebuild need is also coming from some other purpose
(although test only).
Again as Dave said:
"For example for me I do not need kernel modules and I always built in
most of functionality in a bzImage, so when I test a new kernel, rebuild
kdump initrd is useless."
~Pratyush