Re: [PATCH v7 1/7] VFS: Introduce new O_DENY* open flags

[Jeff Layton <jlayton@redhat.com>]

On Fri, 17 Jan 2014 18:18:47 +0000
One Thousand Gnomes <gnomes@lxorguk.ukuu.org.uk> wrote:

> > +#define ESHAREDENIED	258	/* File is locked with a sharelock */
> 
> Have you prepared C library patches to match this ?
> 
> (and why not just use EPERM, it has the meaning you want already)
> 

Tough call...

On the one hand, ESHAREDENIED is a distinct error code so an
application has the ability to determine what happened when an open or
unlink fails.

OTOH, a lot of applications won't understand ESHAREDENIED and may barf
on it. Those apps might handle EPERM better.

I'm not sure what the right approach is there...

> 
> > + * Check to see if there's a share_reservation conflict. LOCK_READ/LOCK_WRITE
> > + * tell us whether the reservation allows other readers and writers.
> > + */
> > +static int
> > +locks_mand_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
> > +{
> 
> Shouldn't this also check for CAP_SYS_DAC or some similar permission so
> that root can override such a mess (eg to fix full disks in an
> emergency) ?
> 
> 

Agreed. This needs a mechanism that allows you to override it, IMO.

CAP_DAC_OVERRIDE doesn't seem quite like the right thing since this
isn't dealing with permissions, per-se. A new capability bit may even be
warranted.

> > +
> > +	/*
> > +	 * For sharelock mounts if a file was created but not opened, we need
> > +	 * to keep parent i_mutex until we finish the open to prevent races when
> > +	 * somebody opens newly created by us file and locks it with a sharelock
> > +	 * before we open it.
> > +	 */
> > +	if (IS_SHARELOCK(dir->d_inode) && error > 0 && *opened & FILE_CREATED) {
> > +		/* Don't check for write permission, don't truncate */
> > +		open_flag &= ~O_TRUNC;
> > +		will_truncate = false;
> > +		acc_mode = MAY_OPEN;
> > +		path_to_nameidata(path, nd);
> > +
> > +		error = may_open(&nd->path, acc_mode, open_flag);
> > +		if (error) {
> > +			mutex_unlock(&dir->d_inode->i_mutex);
> > +			goto out;
> > +		}
> > +		file->f_path.mnt = nd->path.mnt;
> > +		error = finish_open(file, nd->path.dentry, NULL, opened);
> > +		if (error) {
> > +			mutex_unlock(&dir->d_inode->i_mutex);
> > +			if (error == -EOPENSTALE)
> > +				goto stale_open;
> > +			goto out;
> > +		}
> > +		error = sharelock_lock_file(file);
> > +		mutex_unlock(&dir->d_inode->i_mutex);
> > +		if (error)
> > +			goto exit_fput;
> > +		goto opened;
> > +	}
> > +
> >  	mutex_unlock(&dir->d_inode->i_mutex);
> 
> What stops the file system changing mount flags via a remount between
> these two ?
> 
> >  
> >  	if (error <= 0) {
> > @@ -3034,6 +3073,18 @@ finish_open_created:
> >  			goto stale_open;
> >  		goto out;
> >  	}
> > +
> > +	if (IS_SHARELOCK(dir->d_inode)) {
> > +		/*
> > +		 * Lock parent i_mutex to prevent races with sharelocks on
> > +		 * newly created files.
> > +		 */
> > +		mutex_lock(&dir->d_inode->i_mutex);
> > +		error = sharelock_lock_file(file);
> > +		mutex_unlock(&dir->d_inode->i_mutex);
> > +		if (error)
> > +			goto exit_fput;
> > +	}
> >  opened:
> --
> To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html


-- 
Jeff Layton <jlayton@redhat.com>