Re: [PATCH 2/2] f2fs: fix error handling in f2fs_end_enable_verity()

[Eric Biggers <ebiggers@kernel.org>]

On Fri, Mar 05, 2021 at 09:37:26AM +0800, Chao Yu wrote:
> > +cleanup:
> > +	/*
> > +	 * Verity failed to be enabled, so clean up by truncating any verity
> > +	 * metadata that was written beyond i_size (both from cache and from
> > +	 * disk) and clearing FI_VERITY_IN_PROGRESS.
> > +	 */
> > +	truncate_inode_pages(inode->i_mapping, inode->i_size);
> > +	f2fs_truncate(inode);
> 
> Eric,
> 
> Truncation can fail due to a lot of reasons, if we fail in f2fs_truncate(),
> do we need to at least print a message here? or it allows to keep those
> meta/data silently.

I suppose we might as well, although hopefully there will already be a message
for the underlying failure reason too.  Also, f2fs_file_write_iter() has the
same issue too, right?

> One other concern is that how do you think of covering truncate_inode_pages &
> f2fs_truncate with F2FS_I(inode)->i_gc_rwsem[WRITE] lock to avoid racing with
> GC, so that page cache won't be revalidated after truncate_inode_pages().

Yes, that does seem to be needed, due to the way the f2fs garbage collection
works.

- Eric