Re: [Dump-devel] dump and largefiles

[<ty...@va...>]

   Date: Tue, 19 Dec 2000 11:58:42 +0100
   From: Stelian Pop <ste...@al...>

   > Since I work only on purerly userspace and kernel LFS systems 
   > I didn't cared to make it dynamic with an
   > autoconf check in LFS, but it should be easy to add for you.

   Does it hurt to use this patch on a non-LFS system ? Even with
   this patch, it will continue to run correctly on a non-LFS system
   I suppose...

I *really* don't like usage of -D_FILE_OFFSET_BITS=64.  This kind of
libc magic scares me, and it's the compatibility issue (especially with
regards to future versions of glibc, since the glibc developers have
proven themselves less than trustworthy about maintaining backwards
compatibility in my eyes).  

In fact, it's not necessary with e2fsprogs; it already supports 64 bit
i/o, and it's abstracted away behind the unix_io and llseek interfaces
in libext2fs.

   > And this is the patch against e2fsprogs-1.19.

   Ted, will you include this in e2fsprogs-1.20 ? If yes, when do you
   plan to release the new version ?

No, I won't include the patch, because the patch is bogus.  There's a
reason why ext2fs_block_iterate passes BLOCK_FLAG_NO_LARGE to
ext2fs_block_iterate2; that's to avoid surprising old binaries which
were linked the libraries, and weren't expecting to be able to deal with
large filesystems.   In addition, the interface for ext2fs_block_iterate
uses a 32 bit value for the block count variable, which isn't enough for
files which have more than 2**32 blocks.

So ext2fs_block_iterate() will return EXT2_ET_FILE_TOO_BIG for
largefiles, just as the LFS specification requires that open/read/write
return EFBIG unless the file was opened using the O_LARGE flag (or the
open64/read64/write64 interfaces are used).  I treat backwards
compatibility very seriously, and that means not making changes that
might break old binaries.

If your application is known to be 64-bit clean, and won't get surprised
by large files, it should use the ext2fs_block_iterate2 interface.
Hence, a largefile-knowledgeable version dump should not require a
special version of e2fsprogs/libext2fs.  The ext2fs_block_iterate2
interface has been around since version 1.12.

						- Ted