Re: Suspected bug infilesystems (UFS,ADFS,BEFS,BFS,ReiserFS) related to sector_t being unsigned, advice requested

[Mike Fedyk <mfedyk@matchmail.com>]

On Tue, Jan 06, 2004 at 11:43:40PM +0100, Jesper Juhl wrote:
> reiserfs_write_unlock(inode->i_sb); is called at the beginning of the
> function, and as far as I can tell it's matched by a call to
> reiserfs_write_unlock(inode->i_sb); at every potential return point in the
> function, and I see no other locks being taken.

OK, good.

> Besides, since  if (block < 0)  will never be true and
> 	reiserfs_write_unlock(inode->i_sb);
> 	return -EIO;
> will never execute in any case, locking should behave identical to what it
> did before removing the code.
> Locking /seems/ OK to me in this function.
> 
> Also, I did a build of fs/reiserfs/ both with and without the above patch,
> and then did a disassemble of inode.o (objdump -d) and compared the
> generated code for reiserfs_get_block , and the generated code is
> byte-for-byte identical in both cases, which means that gcc realizes that
> the if() statement will never execute and optimizes it away in any case.

I'm not talking about before and afte your patch, I'm talking about before
and after the sector_t patch (presumably for the large block device (gt 2TB)
support).

> I'm not familliar with those "sector_t merges" you are refering to, but I
> found some mention of a 64bit sector_t merge in the 2.5.x kernel
> Changelogs, so I downloaded the 2.5.10 kernel source (first reference to
> sector_t I found was in the 2.5.11 changelog) and took a look at how
> sector_t used to be defined. It seems that it was an unsigned value even
> back then.
> Has sector_t ever been signed?

Really?  Interesting.  Then maybe this is from ported 2.2 code?