Hello! I am working in an environment with an aufs root partition. The top write enabled layer is tmpfs, and the bottom layer is an ext2 partition located on flash media. There are also directories in the filesystem which reside on the same ext2 file system that have been mounted with mount --bind on top of aufs. The following should illustrate this more clearly. device  mountpoint  type [1]root.rw    /rw      tmpfs /dev/sda1  /ro      ext2 root.union  /       aufs /dev/sda1  /usr/configdir  ext2  (<-- this is a mount --bind of /ro/usr/configdir) /dev/sda1  /usr/vardir    ext2  (<-- this is a mount --bind of /ro/usr/vardir) /rw is there mainly for troubleshooting purposes, it is never written to directly. However, there are many scripts and tasks within this system which make changes to /usr/configdir and /usr/vardir. There are also infrequent but occasional changes directly within /ro. The idea is to create a writeable partition which can have many changes made without burning out the write-limited flash media, and which do not persist between reboot. Occasionally, however, changes must be made to the system which do need to persist after reboot, and so we make use of the /ro, /usr/configdir and /usr/vardir directories. The /usr/configdir and /usr/vardir directories are there mainly for backwards compatibility from before we begun using aufs in this manner. Under normal operation, the ext2 filesystem is indeed mounted read-only. When changes need to be made, a script is used which calls mount -o remount,rw /ro. In addition, we call mount -o remount,udba=notify /. Once all of the writes have been made, we then change the /ro mountpoint back to read-only. When we do this, we call auplink first, followed by a sync call, then the remount,ro call. Once we have set the ext2 back to ro, we then mount --remount,udba=reval /. The system is a modified Ubuntu 12.04 system. My latest test branch for working with this problem is using the kernel from Ubuntu Trusty (14.04) git repo. This reports its version as 3.13.0-32-generic. modinfo on the aufs module reports version as 3.13-20140303. The system that this takes place on is remote. When we make version changes to our system, we pull those changes using a script which runs an rsync directly to the /ro mount. before this firmware-upgrade script runs, it ensures the ext2 is mounted ro, and then runs an fsck on it. When this fsck runs, it almost always reports many errors. I will paste some of them into the bottom of this email. Even mounted ro, fsck complains about making changes to a running filesystem when correcting these errors. However, if we force it to make the changes, reboot, and then re-run the firmware-upgrade immediately it will successfully rsync the files to /ro. The firmware-upgrade script then causes a reboot and those changes take effect. During my testing, I reconfigured the unit to not use aufs for its root partition. In order to do this, I had to mount the ext2 as rw for testing. When running fsck -nf /dev/sda1 while /dev/sda1 was still mounted rw, it would have errors. Then, I would mount -o remount,ro /dev/sda1, and run fsck -nf /dev/sda1. This second fsck would not show any errors. In addition, fsck -fy /dev/sda1 did not complain about running checks on a mounted filesystem anymore. During my testing, I also tried reformatting the ext2 partition as ext4, but without a journal. This did not change the results. I am concerned about the recurring filesystem errors which are apparently not being correctly handled by the system when aufs is in use. Please let me know how I can troubleshoot this further, and especially if you need clarification or other information. The errors differ from time to time, but this is one example of the error output: e2fsck 1.42 (29-Nov-2011) /dev/sda1 is mounted.  Pass 1: Checking inodes, blocks, and sizes Deleted inode 229380 has zero dtime.  Fix? yes Pass 2: Checking directory structure Pass 3: Checking directory connectivity /lost+found not found.  Create? yes Pass 3A: Optimizing directories Pass 4: Checking reference counts Pass 5: Checking group summary information Block bitmap differences:  -921603 Fix? yes Free blocks count wrong for group #28 (9828, counted=9829). Fix? yes Free blocks count wrong (753498, counted=753499). Fix? yes Inode bitmap differences:  -229380 Fix? yes Free inodes count wrong for group #28 (5108, counted=5109). Fix? yes Free inodes count wrong (221502, counted=221503). Fix? yes cipa_flash: ***** FILE SYSTEM WAS MODIFIED ***** cipa_flash: 40641/262144 files (0.1% non-contiguous), 294821/1048320 blocks -- Jacob Burkamper CIPAFilter Development Email: [2]jac...@cipafilter.com ---------------------- CIPAFilter Beta Program Email: [3]b...@cipafilter.com Web:     [4]http://www.cipafilter.com
References 1. http://root.rw/ 2. mailto:jac...@cipafilter.com 3. mailto:b...@cipafilter.com 4. http://www.cipafilter.com/
------------------------------------------------------------------------------ Infragistics Professional Build stunning WinForms apps today! Reboot your WinForms applications with our WinForms controls. Build a bridge from your legacy apps to the future. http://pubads.g.doubleclick.net/gampad/clk?id=153845071&iu=/4140/ostg.clktrk
