Re: [PATCH 1/2] Make cramfs little endian only

[Andi Drebes <lists-receive@programmierforen.de>]

> > +/* Converts a cramfs_info from little endian to big endian. */
> > +static inline void cramfs_convert_inode_letobe(struct cramfs_inode* inode)
> > +{
> > +	u8* inode_bytes = (u8*)inode;
> > +	u8 old_nloffs[4];
> > +
> > +	inode->mode = swab16(inode->mode);
> > +	inode->uid = swab16(inode->uid);
> > +	inode->size = (inode_bytes[6] << 16) | (inode_bytes[5] << 8) | (inode_bytes[4]);
> 
> eww.  Is there a nicer way of doing that?
Yes, I know it's ugly. I searched for a swab24 function. The file
fs/jfs/endian24.h defines such a function. Including this file in
fs/cramfs/inode.c would be even more ugly than the code above. What
about moving this to a header file in include/linux/byteorder? (just a suggestion,
perhaps it isn't worth the effort...)

> Might be a bit tricky given the weird way in which struct cramfs_inode was
> defined.
> 
> > +
> > +	/* Save the old values of the namelength and the offset */
> > +	memcpy(old_nloffs, inode_bytes+8, 4);
> > +
> > +	/* Convert the namelength and the offset */
> > +	inode_bytes[8] = ((old_nloffs[0] & 0x3f) << 2) | ((old_nloffs[3] & 0xc0) >> 6);
> > +	inode_bytes[9] = ((old_nloffs[3] & 0x3f) << 2) | ((old_nloffs[2] & 0xc0) >> 6);
> > +	inode_bytes[10] = ((old_nloffs[2] & 0x3f) << 2) | ((old_nloffs[1] & 0xc0) >> 6);
> > +	inode_bytes[11] = ((old_nloffs[1] & 0x3f) << 2) | ((old_nloffs[0] & 0xc0) >> 6);
> > +}
> > +
> > +/* Converts a cramfs superblock from little endian to big endian. */
> > +static inline void cramfs_convert_super_letobe(struct cramfs_super* super)
> > +{
> > +	super->magic = swab32(super->magic);
> > +	super->size = swab32(super->size);
> > +	super->flags = swab32(super->flags);
> > +	super->future = swab32(super->future);
> > +	cramfs_convert_info_letobe(&super->fsid);
> > +	cramfs_convert_inode_letobe(&super->root);
> > +}
> 
> These inlines are not sane.  Just removing those three takes the sparc64
> fs/cramfs/inode.o from 6856 bytes of text down to 5668, which is rather a
> large difference.
OK. I'll remove the inline keywords in my copies aswell. Should I prepare a new patch
and send it to the mailinglist?

> The patch has a number of trivial coding-style errors. 
> scripts/checkpatch.pl finds them.
Oh, I'm sorry. I should have checked that.

	Andi