Hello Theodore,
On Thu, 6 Jan 2000, Theodore Y. Ts'o wrote:
> From: "Manfred Spraul" <[EMAIL PROTECTED]>
> Date: Thu, 6 Jan 2000 20:08:59 +0100
>
> Ok, so the limits are even more complex:
> * on an Alpha, you could build a 4TB ext2 volume and use it.
> * if your Alpha crashes, and you connect your disks to an i386 box, then you
> could corrupt your file system unless someone has added safety checks.
>
> I'm merely try to verify that these safety checks exist:
> eg. ext2 on Alpha supports files > 2 GB, but linux-2.2 on i386 doesn't. If
> you try to mount the disk on i386, then ext2 refuses to perform invalid
> operations on these files.
>
> We're *way* ahead of you. There is a special read-only feature flag,
> EXT2_FEATURE_RO_COMPAT_LARGE_FILE, which is set by a kernel when a file
> is extended past the 2 GB mark. Kernels that don't understand this
> feature flag will refuse to mount the filesystem read-write. Kernels
> that can support large files (which is to say 64 bit platforms and
> 32-bit platforms with the LFS patches applied) will deal with the
> filesystem correctly.
the above are about maximum file size and not filesystem size. Manfred and
others were talking about maximum filesystem size and relevant block
device drivers limitation(s).
> The on-disk format of the ext2 filesystem only allows 4 bytes for the
> block number, and it's defined to be an unsigned value. So if you're
> using 4k blocks, the maximum theoretical filesize is 64 TB. (i.e.,
> 4*2^10*2^32 == 2^46)
again, that is filesize, not filesystem size.
So, it would seem to me that you either did not answer Manfred's question
about what happens if one transfers a large (>2T) ext2 filesystem (with no
large >2G files ever created in it) to i386 or you answered it in a way I
did not understand.
Regards,
Tigran.