Re: e2fsprogs alignment issues

[Eric Sandeen <sandeen@redhat.com>]

On 7/11/12 3:05 PM, Theodore Ts'o wrote:
> On Wed, Jul 11, 2012 at 10:04:51AM -0500, Eric Sandeen wrote:

...

>>  I think Gordan suggested (if I understand it
>> right) that doing an array of ints might also solve the problem, since
>> ints should be on natural alignment.  Or maybe in some cases malloc/free
>> would be more obvious, if handling errors isn't too tricky.
> 
> In the specific case which Gordon has pointed out, the obvious thing
> to do is to just to set errno to ENOMEM, and return -1. since we
> already reflect an error code up to the caller if the FIEMAP ioctl()
> fails.
> 
> If someone sends me the patch, I will happily apply it.

Sure, I was planning on it :)

>> (IIRC "make gcc-wall" will also emit warnings for casts which change
>> natural alignment, among other things)
> 
> I'd have to check to be sure, but I don't think so, since it would
> have way too many false positives.  We *do* have code where we take
> char *'s and and then cast them to some other pointer type, and then
> dereference them.  And we do currently assume that it is safe to do
> this for on-disk data structures which are 4 byte aligned, in the
> directory entry code, for example.
> 
> I will *not* accept a patch which uses memcpy to copy each field in
> the on-disk superblock, or directory entry, into an int, just in case
> there is some insane architecture which requires that 4 byte integers
> be 32-byte aligned, or something else insane like that.

Well, let's just see where we're at, first, and see what it'll take,
case by case.

-Eric