Re: ctime set by truncate even if NOCMTIME requested

[Trond Myklebust <trond.myklebust@fys.uio.no>]

ty den 20.09.2005 Klokka 12:05 (+0200) skreiv Miklos Szeredi:
> These are othogonal problems.
> 
> IS_NOCMTIME is the filesystem's way of saying that it doesn't need
> ->setattr on truncate(), write(), etc.  Why?  Because it can do the
> [cm]time change implicitly _within_ the operation.

No. IS_NOCMTIME is the filesystem's way of telling the VFS never to
screw around with the values of inode->i_mtime and inode->i_ctime. 

The reason is that crap like inode_update_time() explicitly sets these
values to the local current time instead of using server timestamps.

OTOH, ->setattr with an ATTR_MTIME or ATTR_CTIME argument is telling the
filesystem to update the timestamp. The filesystem then has a choice of
whether or not to use current time, server time, or to just ignore it if
the timestamp is going to be be updated by the other ->setattr arguments
anyway.

> ATTR_MTIME is _only_ set in utime[s], which all filesystems want to
> honor.

ATTR_MTIME is set in both utimes and truncate. In the latter case, CIFS
could optimise it away by noting that ATTR_SIZE will set mtime anyway.
As for ATTR_CTIME, that is also set in chown(), chmod(). It too can be
optimised away for those operations, assuming that CIFS servers
automatically update ctime.

Cheers,
  Trond