Re: [PATCH] ext4: Use page_mkwrite vma_operations to get mmap write notification.

[Mingming Cao <cmm@us.ibm.com>]

On Fri, 2008-02-22 at 20:09 +0530, Aneesh Kumar K.V wrote:
> We would like to get notified when we are doing a write on mmap section.
> This is needed with respect to preallocated area. We split the preallocated
> area into initialzed extent and uninitialzed extent in the call back. This
> let us handle ENOSPC better. Otherwise we get ENOSPC in the writepage and
> that would result in data loss. The changes are also needed to handle ENOSPC
> when writing to an mmap section of files with holes.
> 
> Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
> ---
>  fs/ext4/file.c          |   19 ++++++++++++++-
>  fs/ext4/inode.c         |   60 +++++++++++++++++++++++++++++++++++++++++++++++
>  include/linux/ext4_fs.h |    1 +
>  3 files changed, 79 insertions(+), 1 deletions(-)
> 
> diff --git a/fs/ext4/file.c b/fs/ext4/file.c
> index 20507a2..77341c1 100644
> --- a/fs/ext4/file.c
> +++ b/fs/ext4/file.c
> @@ -123,6 +123,23 @@ force_commit:
>  	return ret;
>  }
> 
> +static struct vm_operations_struct ext4_file_vm_ops = {
> +	.fault		= filemap_fault,
> +	.page_mkwrite   = ext4_page_mkwrite,
> +};
> +
> +static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
> +{
> +	struct address_space *mapping = file->f_mapping;
> +
> +	if (!mapping->a_ops->readpage)
> +		return -ENOEXEC;
> +	file_accessed(file);
> +	vma->vm_ops = &ext4_file_vm_ops;
> +	vma->vm_flags |= VM_CAN_NONLINEAR;
> +	return 0;
> +}
> +
>  const struct file_operations ext4_file_operations = {
>  	.llseek		= generic_file_llseek,
>  	.read		= do_sync_read,
> @@ -133,7 +150,7 @@ const struct file_operations ext4_file_operations = {
>  #ifdef CONFIG_COMPAT
>  	.compat_ioctl	= ext4_compat_ioctl,
>  #endif
> -	.mmap		= generic_file_mmap,
> +	.mmap		= ext4_file_mmap,
>  	.open		= generic_file_open,
>  	.release	= ext4_release_file,
>  	.fsync		= ext4_sync_file,
> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> index 5b5d63d..00af97d 100644
> --- a/fs/ext4/inode.c
> +++ b/fs/ext4/inode.c
> @@ -3490,3 +3490,63 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
> 
>  	return err;
>  }
> +
> +int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
> +{
> +	unsigned long end;
> +	loff_t size;
> +	handle_t *handle;
> +	int ret = -EINVAL, needed_blocks;
> +	struct file *file   = vma->vm_file;
> +	struct inode *inode = file->f_path.dentry->d_inode;
> +
> +	needed_blocks = ext4_writepage_trans_blocks(inode);
> +	/* We need to take inode mutex to prevent parallel write */
> +	mutex_lock(&inode->i_mutex);
> +	lock_page(page);
> +	size = i_size_read(inode);
> +	if ((page->mapping != inode->i_mapping) ||
> +	    (page_offset(page) > size)) {
> +		/* page got truncated out from underneath us */
> +		goto out_unlock;
> +	}
> +
> +	/* page is wholly or partially inside EOF */
> +	if (((page->index + 1) << PAGE_CACHE_SHIFT) > size)
> +		end = size & ~PAGE_CACHE_MASK;
> +	else
> +		end = PAGE_CACHE_SIZE;
> +
> +	handle = ext4_journal_start(inode, needed_blocks);
> +	if (IS_ERR(handle)) {
> +		ret = PTR_ERR(handle);
> +		goto out_unlock;
> +	}
> +	/* Will zero out the pages if buffer is marked new */
> +	ret = block_prepare_write(page, 0, end, ext4_get_block);
> +
> +	if (!ret && ext4_should_journal_data(inode)) {
> +		ret = walk_page_buffers(handle, page_buffers(page),
> +				0, end, NULL, do_journal_get_write_access);
> +		if (!ret)
> +			ret = walk_page_buffers(handle, page_buffers(page),
> +						0, end, NULL, write_end_fn);
> +		/*
> +		 * we don't want to call block_commit_write in journalled mode
> +		 */
> +		ext4_journal_stop(handle);
> +		goto out_unlock;
> +	}
> +	if (!ret && ext4_should_order_data(inode)) {
> +		ret = walk_page_buffers(handle, page_buffers(page),
> +					0, end, NULL, ext4_journal_dirty_data);
> +	}
> +	if (!ret)
> +		ret = block_commit_write(page, 0, end);
> +
Hmm, it seems wired to do commit_write when the page is about becoming
writable, but maybe that's the way it needs to?

Don't we need to update the i_size somewhere?

> +	ext4_journal_stop(handle);
> +out_unlock:
> +	unlock_page(page);
> +	mutex_unlock(&inode->i_mutex);
> +	return ret;
> +}

It seems this combined the three journalling mode prepare_write() code
here:( 

Since prepare_write() and commit_write() is going to sunset, why not
simply calling mappings->a_ops->write_begin() and then write_end()? that
should take care of pretty much the journalling and the page operations,
no?

Mingming
> diff --git a/include/linux/ext4_fs.h b/include/linux/ext4_fs.h
> index 22810b1..8f5a563 100644
> --- a/include/linux/ext4_fs.h
> +++ b/include/linux/ext4_fs.h
> @@ -1059,6 +1059,7 @@ extern void ext4_set_aops(struct inode *inode);
>  extern int ext4_writepage_trans_blocks(struct inode *);
>  extern int ext4_block_truncate_page(handle_t *handle, struct page *page,
>  		struct address_space *mapping, loff_t from);
> +extern int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page);
> 
>  /* ioctl.c */
>  extern long ext4_ioctl(struct file *, unsigned int, unsigned long);