Description of problem: /sbin/mkdumprd only includes the pieces needed for scp if a net: type destination is found /etc/kdump.conf. As a result, 'kcp --local /proc/vmcore /mnt/var...' fails because the execve of 'scp' fails. Version-Release number of selected component (if applicable): kexec-tools-1.101-51.fc6 How reproducible: Create a /etc/kdump.conf that only contains: ext3 /dev/something_valid default reboot Steps to Reproduce: 1. 2. 3. Actual results: kcp creates the directory correctly. It then gets to the execlp at kcp.c:168 and dies because scp and its requisites weren't included in the initrd. Expected results: Additional info: In some environments, you may need to take crash dumps locally, without talking to the network (stand-alone machines, or even machines on the net that don't have a valid place to NFS or SCP a large file to)... In addition, the "dest" parameter in the kcp.c code is a directory spec, which is just a bug waiting to happen to somebody who tries to provide their own 'scp' that just does a local copy (you need to sprintf(new,"%s/vmcore",dest); or similar to get a filename useful for writing (as I discovered the hard way while trying to create a local workaround.. ;)
I've re-written kexec-tools to use busybox, so kcp is no longer needed. ssh should be included in that initrd as well. Version -53.fc6 has all these changes