Re: [PATCH 1/3] mm/filemap: initialize fsdata with iocb->ki_flags

[Christoph Hellwig <hch@infradead.org>]

On Wed, May 14, 2025 at 02:38:05PM +0100, Matthew Wilcox wrote:
> On Tue, May 13, 2025 at 11:51:25PM -0400, Theodore Ts'o wrote:
> > I understand that it would be a lot more inconvenient change the
> > function signature of write_begin() to pass through iocb->ki_fags via
> > a new parameter.  But I think that probably is the best way to go.
> 
> I'd suggest that passing in iocb rather than file is the way to go.
> Most callers of ->write_begin already pass NULL as the first argument so
> would not need to change.  i915/gem passes a non-NULL file, but it only
> operates on shmem and shmem does not use the file argument, so they can
> pass NULL instead.

i915/gem needs to stop using write_begin/end and just do an iter_write.
Someone who has the hardware and/or CI setup needs to figure out if
vfs_write and kernel_write is fine, or this is magic enough to skip
checks and protection and go straight to callig ->iter_write.

> fs/buffer.c simply passes through the file passed
> to write_begin and can be changed to pass through the iocb passed in.
> exfat_extend_valid_size() has an iocb in its caller and can pass in the
> iocb instead.  generic_perform_write() has an iocb.

And we really need to stop theading this through the address_space
ops because it's not a generic method but a callback for the file
system..