Re: [PATCH v3 3/4] fs: change write_begin/write_end interface to take struct kiocb *

[Taotao Chen <chentt325@gmail.com>]

在 2025/6/27 23:45, Matthew Wilcox 写道:
> On Fri, Jun 27, 2025 at 11:03:11AM +0000, 陈涛涛 Taotao Chen wrote:
>> diff --git a/fs/exfat/file.c b/fs/exfat/file.c
>> index 841a5b18e3df..fdc2fa1e5c41 100644
>> --- a/fs/exfat/file.c
>> +++ b/fs/exfat/file.c
>> @@ -532,10 +532,12 @@ int exfat_file_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
>>   	return blkdev_issue_flush(inode->i_sb->s_bdev);
>>   }
>>   
>> -static int exfat_extend_valid_size(struct file *file, loff_t new_valid_size)
>> +static int exfat_extend_valid_size(const struct kiocb *iocb,
>> +				   loff_t new_valid_size)
>>   {
>>   	int err;
>>   	loff_t pos;
>> +	struct file *file = iocb->ki_filp;
>>   	struct inode *inode = file_inode(file);
>>   	struct exfat_inode_info *ei = EXFAT_I(inode);
>>   	struct address_space *mapping = inode->i_mapping;
>> @@ -551,14 +553,14 @@ static int exfat_extend_valid_size(struct file *file, loff_t new_valid_size)
>>   		if (pos + len > new_valid_size)
>>   			len = new_valid_size - pos;
>>   
>> -		err = ops->write_begin(file, mapping, pos, len, &folio, NULL);
>> +		err = ops->write_begin(iocb, mapping, pos, len, &folio, NULL);
>>   		if (err)
>>   			goto out;
>>   
>>   		off = offset_in_folio(folio, pos);
>>   		folio_zero_new_buffers(folio, off, off + len);
>>   
>> -		err = ops->write_end(file, mapping, pos, len, len, folio, NULL);
>> +		err = ops->write_end(iocb, mapping, pos, len, len, folio, NULL);
>>   		if (err < 0)
>>   			goto out;
>>   		pos += len;
>> @@ -604,7 +606,7 @@ static ssize_t exfat_file_write_iter(struct kiocb *iocb, struct iov_iter *iter)
>>   	}
>>   
>>   	if (pos > valid_size) {
>> -		ret = exfat_extend_valid_size(file, pos);
>> +		ret = exfat_extend_valid_size(iocb, pos);
>>   		if (ret < 0 && ret != -ENOSPC) {
>>   			exfat_err(inode->i_sb,
>>   				"write: fail to zero from %llu to %llu(%zd)",
>> @@ -655,8 +657,11 @@ static vm_fault_t exfat_page_mkwrite(struct vm_fault *vmf)
>>   	struct file *file = vma->vm_file;
>>   	struct inode *inode = file_inode(file);
>>   	struct exfat_inode_info *ei = EXFAT_I(inode);
>> +	struct kiocb iocb;
>>   	loff_t start, end;
>>   
>> +	init_sync_kiocb(&iocb, file);
>> +
>>   	if (!inode_trylock(inode))
>>   		return VM_FAULT_RETRY;
>>   
>> @@ -665,7 +670,7 @@ static vm_fault_t exfat_page_mkwrite(struct vm_fault *vmf)
>>   			start + vma->vm_end - vma->vm_start);
>>   
>>   	if (ei->valid_size < end) {
>> -		err = exfat_extend_valid_size(file, end);
>> +		err = exfat_extend_valid_size(&iocb, end);
>>   		if (err < 0) {
>>   			inode_unlock(inode);
>>   			return vmf_fs_error(err);
> This is unnecessary work.  The only ->write_begin/write_end that we'll
> see here is exfat_write_begin() / exfat_write_end() which don't actually
> need iocb (or file).  Traditionally we pass NULL in these situations,
> but the exfat people probably weren't aware of this convention.
>
> exfat_extend_valid_size() only uses the file it's passed to get the
> inode, and both callers already have the inode.  So I'd change
> exfat_extend_valid_size() to take an inode instead of a file as its
> first argument, then you can skip the creation of an iocb in
> exfat_page_mkwrite().


My initial goal was to maintain consistency with the updated ->write_begin/

->write_end interfaces. That meant passing the iocb to avoid special cases

and keep the changes minimal and safe.

But you're right, since exfat_write_begin() and exfat_write_end() don't 
use the

iocb or file pointer, passing NULL is fine, and having 
exfat_extend_valid_size()

directly take an inode makes the code simpler and clearer.


In addition, inside the ntfs_extend_initialized_size() function, I also 
set the iocb

parameter to NULL when calling ntfs_write_begin() and ntfs_write_end().