
Dear Tom Rini,
On 09/12/2012 03:57 PM, Rob Herring wrote:
On 09/12/2012 05:49 PM, Tom Rini wrote:
On Mon, Aug 13, 2012 at 01:28:19PM -0500, Rob Herring wrote:
On 08/13/2012 06:52 AM, Wolfgang Denk wrote:
Dear Rob Herring,
In message 50244D5A.3080304@gmail.com you wrote:
I reported already that the prior version that ext4 has issues with sub-directories. I don't think that has been addressed in V5. Some directories show up fine and some don't. So it's kind of random whether u-boot can read a /boot directory. This was after full ubuntu installs. I'd guess a simple test with a couple of files and directories will not show the problem.
We really need a test case here. In my (certainly not very extensive) tests I didn't see such a problem.
Can you please describe what failed for you?
I do an ubuntu install to a single ext4 fs and then ext2ls gives this:
[snip]
The problem is in the directories with sizes of 0. It does seem to be directories with higher blkno's. Perhaps the lack of support for hash table directory entries is a problem. Just guessing here as I don't know much about ext4 structure.
I haven't been able to produce a simple example just creating a bunch of files and directories, so only the disk after an ubuntu install has the problem.
I took an Ubuntu install I had (for x86) that was on a USB drive, had been pretty extensively used, and was is ext4. I couldn't find any size 0 directories. Are you able to reproduce this problem on other hardware? Is the image in question NOT remountable as ext3 (extents is set, generated) ? My biggest concern is breakage among ext2/3 filesystems. Thanks!
What size partition? It is also fine for me with a small 1.2GB partition. I only see the problem with larger partitions (most of a 250GB drive).
This is a 240G partition.
I was under the impression that ext4 is not mountable as ext3.
You can go ext3->"ext4"->ext3, it's only when you have fsck rebuild and make use of extents, iirc, that it's no longer mountable as ext3.
Maybe you can pick that loop block driver from Pavel [1] and try it on sandbox target? That might ease testing etc.
[1] http://patchwork.ozlabs.org/patch/182160/
Best regards, Marek Vasut