On Wed, Feb 07, 2007 at 10:18:23AM +0000, Christoph Hellwig wrote:
> This looks to me. But given that this is generic code except for the
> get_block callback, shouldn't we put the guts into buffer.c and wire
> all filesystems up to use it? e.g.
> int block_page_mkwrite(struct vm_area_struct *vma, struct page *page,
> get_block_t get_block)
> struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
> unsigned long end;
> int ret = 0;
> if ((page->index + 1) << PAGE_CACHE_SHIFT > i_size_read(inode))
> end = i_size_read(inode) & ~PAGE_CACHE_MASK;
> end = PAGE_CACHE_SIZE;
> ret = block_prepare_write(page, 0, end, block);
> if (!ret)
> ret = block_commit_write(page, 0, end);
> return ret;
> and then in xfs and similar in other filesystems:
> STATIC int
> struct vm_area_struct *vma,
> struct page *page)
> return block_page_mkwrite(vma, page, xfs_get_blocks);
Yes, that can be done. block_page_mkwrite() would then go into
fs/buffer.c? My patch originally had a bunch of other stuff and
i wasn't sure that it could be done with generic code.
I'll send an updated patch in a little while.
> BTW, why is xfs_get_blocks not called xfs_get_block?
I presume because it replaced the xfs_get_block() function when the
block mapping callouts were modified to support mapping of multiple
blocks. Maybe you should ask Nathan that question. ;)
SGI Australian Software Group