On 29/07/14 10:46, Theodore Ts'o wrote:
On Sat, Jul 26, 2014 at 09:46:20PM +0800, Brad Campbell wrote:
Fast forward 12 months and
I've added 2 drives to the array and bumped up against this issue. So it was
initially 4883458240 blocks. It would have been created with e2fsprogs from
Debian Stable (so 1.42.5).
Confirmed; debian stable's mke2fs version 1.42.5 does the wrong thing
if you do:
mke2fs -t ext4 -O 64bit,resize_inode /mnt/foo.img 19T
It creates a file system which is has a resize_inode which --- well,
is in a format that the kernel isn't capable of handling and which is
not something that was originally intended. The interesting thing is
that resize2fs 1.42.11 is apparently able to handle resizing the above
file system created with mke2fs 1.42.5.
And it did just fine once patched for the 32bit overflow. The off-line
resize went without a hitch. Oddly enough the Kernel took it out to
5641863168 blocks ok in the first online attempt.
Mke2fs 1.42.12 will avoid creating a resize_inode for file systems >
16T, so this situation doesn't arise.
I need to take a closer look to see if there's some way I can teach
the kernel to be able to deal with this file system, but for now, you
can work around it by using tune2fs to remove the resize_inode, and
after running e2fsck and remounting the file system, the on-line
resize will work correctly.
Would it not be easier just to put a test in resize2fs which says "I
can't do that, please fix your filesystem first" ?
1.42.5 won't try (32 bit limit).
1.42.11 will spin (32 bit overflow)
So just make 1.42.12 abort with an error.
Is it worth having e2fsck check for mis-applied features and flags?
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
