Re: [PATCH 2.5.74-mm2] nbd: make nbd and block layer agree about device and block sizes

[Jeff Garzik <jgarzik@pobox.com>]

Paul Clements wrote:
>  	case NBD_SET_BLKSIZE:
> -		if ((arg & (arg-1)) || (arg < 512) || (arg > PAGE_SIZE))
> -			return -EINVAL;
>  		lo->blksize = arg;
> -		lo->bytesize &= ~(lo->blksize-1); 
> +		lo->bytesize &= ~(lo->blksize-1);
> +		inode->i_bdev->bd_inode->i_size = lo->bytesize;
> +		set_blocksize(inode->i_bdev, lo->blksize);
>  		set_capacity(lo->disk, lo->bytesize >> 9);
>  		return 0;
>  	case NBD_SET_SIZE:
> -		lo->bytesize = arg & ~(lo->blksize-1); 
> +		lo->bytesize = arg & ~(lo->blksize-1);
> +		inode->i_bdev->bd_inode->i_size = lo->bytesize;
> +		set_blocksize(inode->i_bdev, lo->blksize);
>  		set_capacity(lo->disk, lo->bytesize >> 9);
>  		return 0;
>  	case NBD_SET_SIZE_BLOCKS:
>  		lo->bytesize = ((u64) arg) * lo->blksize;
> +		inode->i_bdev->bd_inode->i_size = lo->bytesize;
> +		set_blocksize(inode->i_bdev, lo->blksize);
>  		set_capacity(lo->disk, lo->bytesize >> 9);
>  		return 0;
>  	case NBD_DO_IT:


Thanks.  You forgot to check set_blocksize's return value for errors, 
though.

Also, I wonder if you found a bug/oversight in set_blocksize -- it sets 
bd_inode->i_blkbits but not bd_inode->i_size.  I think it should set 
i_size also.  Maybe Andrew or Al can confirm/refute this assertion.

	Jeff