On Aug 5, 2012 6:04 AM, "C Anthony Risinger" <anthony@xxxxxxx> wrote: > > On Sat, Aug 4, 2012 at 7:53 AM, Jackson Alley <toomanymirrors@xxxxxxxxx> wrote: > > > > I've rebooted and confirmed the udev rule I found does not resolve the > > issue. I don't believe the problem lies with mkinitcpio as I'm not using > > a root btrfs system but rather in the init scripts > > ( https://projects.archlinux.org/initscripts.git/commit/?id=13ca7f028ac775773a6685966d0faeee5cc55a9d ). > > I've attached the boot log and dmesg output. > > dmesg: http://pastebin.com/PxfnYgAm > > boot log: http://pastebin.com/Pg9xY4u5 > > yeah i didn't think mkinitcpio was involved but you said it was > throwing an error about a missing btrfs hook, even though you have no > hook defined ... so i asked. > > AFIAK the FS module is not loaded automatically until mount time -- > dmesg shows the module being loaded at that time, not when scanned. > dracut loads the module just prior to calling `device scan`: > > http://git.kernel.org/?p=boot/dracut/dracut.git;a=blob;f=modules.d/90btrfs/80-btrfs.rules;h=e74f9a6af647e465d9367e95b37aef2e57ddfdbe;hb=HEAD > > ... and since scanning requires /dev/btrfs-control, which is not > created until the btrfs module is loaded, the module MUST be loaded to > properly register the device. i was able to confirm that `btrfs > device scan` DOES NOT automatically load the required `btrfs` module. > > your dmesg shows required devices loading at this time: > > [ 4.058562] sd 9:0:0:0: [sde] 976773168 512-byte logical blocks: > (500 GB/465 GiB) > > ... but the module is not loaded until after all devices fire: > > [ 4.171028] Btrfs loaded > > ... therefore probably nothing was registered, and thus a few seconds > later you see: > > [ 6.081051] btrfs: failed to read the system array on sde1 > > ... try this (tested for one device, unverified for more): > > # cat /usr/lib/udev/rules.d/70-btrfs.rules > ACTION!="remove", SUBSYSTEM=="block", ENV{ID_FS_TYPE}=="btrfs", > RUN+="/sbin/modprobe btrfs", RUN+="/usr/bin/btrfs device scan > $env{DEVNAME}" This makes a lot of sense. Please verify. Tom
