Re: [PATCH 2/6] vfs: flush and wait for io when setting the immutable flag via SETFLAGS

[Darrick J. Wong]

On Thu, Jun 20, 2019 at 04:00:28PM +0200, Jan Kara wrote:
> On Mon 10-06-19 21:46:25, Darrick J. Wong wrote:
> > From: Darrick J. Wong <darrick.wong@xxxxxxxxxx>
> > 
> > When we're using FS_IOC_SETFLAGS to set the immutable flag on a file, we
> > need to ensure that userspace can't continue to write the file after the
> > file becomes immutable.  To make that happen, we have to flush all the
> > dirty pagecache pages to disk to ensure that we can fail a page fault on
> > a mmap'd region, wait for pending directio to complete, and hope the
> > caller locked out any new writes by holding the inode lock.
> > 
> > Signed-off-by: Darrick J. Wong <darrick.wong@xxxxxxxxxx>
> 
> ...
> 
> > diff --git a/fs/ext4/ioctl.c b/fs/ext4/ioctl.c
> > index 6aa1df1918f7..a05341b94d98 100644
> > --- a/fs/ext4/ioctl.c
> > +++ b/fs/ext4/ioctl.c
> > @@ -290,6 +290,9 @@ static int ext4_ioctl_setflags(struct inode *inode,
> >  	jflag = flags & EXT4_JOURNAL_DATA_FL;
> >  
> >  	err = vfs_ioc_setflags_check(inode, oldflags, flags);
> > +	if (err)
> > +		goto flags_out;
> > +	err = vfs_ioc_setflags_flush_data(inode, flags);
> >  	if (err)
> >  		goto flags_out;
> >  
> 
> ...
> 
> > diff --git a/include/linux/fs.h b/include/linux/fs.h
> > index 8dad3c80b611..9c899c63957e 100644
> > --- a/include/linux/fs.h
> > +++ b/include/linux/fs.h
> > @@ -3548,7 +3548,41 @@ static inline struct sock *io_uring_get_socket(struct file *file)
> >  
> >  int vfs_ioc_setflags_check(struct inode *inode, int oldflags, int flags);
> >  
> > +/*
> > + * Do we need to flush the file data before changing attributes?  When we're
> > + * setting the immutable flag we must stop all directio writes and flush the
> > + * dirty pages so that we can fail the page fault on the next write attempt.
> > + */
> > +static inline bool vfs_ioc_setflags_need_flush(struct inode *inode, int flags)
> > +{
> > +	if (S_ISREG(inode->i_mode) && !IS_IMMUTABLE(inode) &&
> > +	    (flags & FS_IMMUTABLE_FL))
> > +		return true;
> > +
> > +	return false;
> > +}
> > +
> > +/*
> > + * Flush file data before changing attributes.  Caller must hold any locks
> > + * required to prevent further writes to this file until we're done setting
> > + * flags.
> > + */
> > +static inline int inode_flush_data(struct inode *inode)
> > +{
> > +	inode_dio_wait(inode);
> > +	return filemap_write_and_wait(inode->i_mapping);
> > +}
> > +
> > +/* Flush file data before changing attributes, if necessary. */
> > +static inline int vfs_ioc_setflags_flush_data(struct inode *inode, int flags)
> > +{
> > +	if (vfs_ioc_setflags_need_flush(inode, flags))
> > +		return inode_flush_data(inode);
> > +	return 0;
> > +}
> > +
> 
> But this is racy at least for page faults, isn't it? What protects you
> against write faults just after filemap_write_and_wait() has finished?
> So either you need to set FS_IMMUTABLE_FL before flushing data or you need
> to get more protection from the fs than just i_rwsem. In the case of ext4
> that would be i_mmap_rwsem but other filesystems don't have equivalent
> protection...

Yes, I see that now.  I think it'll work to set S_IMMUTABLE before
trying the flush, so long as I am careful to put the call sites right
before we update the inode flags.

--D

> 
> 								Honza
> -- 
> Jan Kara <jack@xxxxxxxx>
> SUSE Labs, CR