I asked around about the usage and requirement of the *scan-commandos. According to the ones on the lvm-mailinglist pvscan and vgscan only has meaning if /etc/lvm/cache needs to be updated, however if /etc is mounted readonly said update cannot occur, and thus at least the pvscan in /lib/rcscripts/lvm-start is unecessary. The only need for vgscan as such seems to be needed if the interaction between your initramfs/initrd and system is broken in such ways that the dev-nodes are not moved from the initramfs/initrd to the real root (genkernel are being worked on to handle this, but is currently broken). In those cases --mknodes are needed to recreate those nodes. Also, the execution of the lvm commands is much faster if we launch them as a lvm-script, since lvm2 in those cases caches steps between commands, and thus does not redo things from the vgscan-step in the vgchange-step. As such I suggest the following: 1. Drop pvscan from /lib/rcscripts/addons/lvm-start 2. Make that script run the lvm commands as a lvm-command, i.e.: lvm_commands="#! /bin/lvm" lvm_commands="${lvm_commands}\nvgscan --mknodes" lvm_commands="${lvm_commands}\nvgchange -ay --sysinit" printf "%b\n" "${lvm_commands}" | /bin/lvm /proc/self/fd/0 In the previous command, the /proc* thing is needed because if lvm is compiled without readline it does not read from a pipe, only from a file, and stdin seems to work fine. Even if we run all three steps (i.e. pvscan, vgscan and vgchange) it is still more then twice as fast running the commands in lvm instead of each step alone, so that change I think should be made even if there is a reason to hold pvscan.
InCVS. I've kept the pvscan, because I've seen cases in the past where the vgscan after it didn't pick up devices.
