Re: re-enable IOCB_NOWAIT writes to files

[Jeff Layton <jlayton@kernel.org>]

On Fri, 2025-11-14 at 16:28 +0100, Christoph Hellwig wrote:
> On Fri, Nov 14, 2025 at 09:04:58AM -0500, Jeff Layton wrote:
> > This all looks pretty reasonable to me. There are a few changelog and
> > subject line typos, but the code changes look fine. You can add:
> 
> Please tell me about them so I can fix them. 
> 
> > As far as nfsd's usage of FMODE_NOCMTIME, it looks OK to me. That's
> > implemented today by the check in file_modified_flags(), which is
> > generic and should work across filesystems.
> 
> Nothing requires file_update_time / file_modified_flags are helpers
> that a file system may or may not call.  I've not done an audit
> if everyone actually uses them.

FWIW, I turned claude loose on this, and it produced:

-------------------------------8<----------------------------------
  Findings:                                                           
                                                                                                         
  1. coda_file_write_iter (fs/coda/file.c:66-94)                      
                                                                                                                                                                                                                   
  Location: fs/coda/file.c:86Issue: Manually updates ctime in
write_iter operation                                                  
  inode_set_mtime_to_ts(coda_inode,
inode_set_ctime_current(coda_inode));
  Context: This is a stacking filesystem that delegates writes to a
container file via vfs_iter_write(), then manually copies attributes
back. However, it bypasses file_update_time() or file_modified(),     
  which means it doesn't handle:                    
  - Read-only filesystem checks                                       
  - Immutable inode checks                                            
  - S_NOCMTIME flag                                                   
  - i_version updates                                                 
  - Proper dirty marking                                              
                                                                                                         
  2. efivarfs_file_write (fs/efivarfs/file.c:15-77)                   

  Location: fs/efivarfs/file.c:66Issue: Manually updates ctime in write
operation                        
  inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
  Context: Uses the legacy .write operation (not .write_iter), but
still a write path that should use proper timestamp helpers.

  3. ocfs2_write_end_nolock (fs/ocfs2/aops.c:1926-2050)

  Location: fs/ocfs2/aops.c:2024-2026Issue: Manually updates ctime in
write_end callback
  inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
  di->i_mtime = di->i_ctime = cpu_to_le64(inode_get_mtime_sec(inode));
  di->i_mtime_nsec = di->i_ctime_nsec =
cpu_to_le32(inode_get_mtime_nsec(inode));
  Context: This is called from OCFS2's write path. While
ocfs2_file_write_iter calls __generic_file_write_iter, the write_end
callback manually manages timestamps. This is used both by regular
writes and
  page_mkwrite via __ocfs2_page_mkwrite.

  4. ubifs_vm_page_mkwrite (fs/ubifs/file.c:1493-1580)

  Location: fs/ubifs/file.c:1570Issue: Manually updates ctime in
page_mkwrite operation
  inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
  Context: UBIFS implements its own timestamp update logic with
mctime_update_needed() and manual budgeting for space, bypassing
file_update_time().
-------------------------------8<----------------------------------

Only ocfs2 is exportable, so I think we want to convert that one. The
others I'm not sure of yet.
-- 
Jeff Layton <jlayton@kernel.org>