On Thu, Sep 06, 2018 at 05:56:31PM +0530, Souptick Joarder wrote: > On Thu, Sep 6, 2018 at 2:08 PM Jan Kara <jack@xxxxxxx> wrote: > > Yes, I'd start with converting ext4_page_mkwrite() - that should be pretty > > straightforward - and we can leave block_page_mkwrite() as is for now. I > > don't think allocating other VM_FAULT_ codes is going to cut it as > > generally the filesystem may need to communicate different error codes back > > and you don't know in advance which are interesting. > > Then I need to take care of ext4_page_mkwrite() and ext4_filemap_fault() > to migrate to use vm_fault_t return type. Everything else can be removed > from this patch and it will go as a separate patch. > > As block_page_mkwrite() is getting called from 2 places in ext4 and nilfs and > both places fault handler code convert errno to VM_FAULT_CODE using > block_page_mkwrite_return(), is it required to migrate block_page_mkwrite() > to use vm_fault_t return type and further complicate the API or better to > leave this API in current state ?? Leave block_page_mkwrite() alone. Somebody who understands it better than you do can take care of converting it, if that's even the right thing to do. Let's get the typedef conversion _finished_ so we get the benefit of typechecking for driver writers.
