On Sat, Dec 06, 2025 at 12:52:37PM +0000, Matthew Wilcox wrote:
> On Sat, Dec 06, 2025 at 01:03:35PM +0100, Ahelenia Ziemiańska wrote:
> > diff --git a/mm/filemap.c b/mm/filemap.c
> > index ebd75684cb0a..0b0d5cfbcd44 100644
> > --- a/mm/filemap.c
> > +++ b/mm/filemap.c
> > @@ -3945,12 +3945,18 @@ EXPORT_SYMBOL(filemap_map_pages);
> >  
> >  vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
> >  {
> > -	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
> > +	struct file *file = vmf->vma->vm_file;
> > +	struct address_space *mapping = file->f_mapping;
> >  	struct folio *folio = page_folio(vmf->page);
> >  	vm_fault_t ret = VM_FAULT_LOCKED;
> >  
> > +	if (unlikely(IS_IMMUTABLE(file_inode(file)))) {
> > +		ret = VM_FAULT_SIGBUS;
> > +		goto out;
> I don't believe you tested this code path.  It contains a rather obvious
> bug.
You're right, I tested this on ext4 twice and didn't realise;
this ought just be return VM_FAULT_SIGBUS;.

But even this doesn't work, because shmem{,_anon}_vm_ops
don't have a page_mkwrite callback at all,
much less it being filemap_page_mkwrite().

I'm not sure how I concluded that they do.

Adding a shmem_page_mkwrite() works for the mmap case.