Re: [PATCH v2] f2fs: fix to disallow getting inner inode via f2fs_iget()

[Chao Yu]

On 2022/9/13 13:54, Jaegeuk Kim wrote:
> On 09/13, Chao Yu wrote:
> > On 2022/9/12 23:39, Jaegeuk Kim wrote:
> > > On 09/08, Chao Yu wrote:
> > > > On 2022/9/8 10:19, Jaegeuk Kim wrote:
> > > > > On 09/08, Chao Yu wrote:
> > > > > > On 2022/9/8 10:02, Jaegeuk Kim wrote:
> > > > > > > On 08/31, Chao Yu wrote:
> > > > > > > > From: Chao Yu <chao.yu@xxxxxxxx>
> > > > > > > >
> > > > > > > > Introduce f2fs_iget_inner() for f2fs_fill_super() to get inner inode:
> > > > > > > > meta inode, node inode or compressed inode, and add f2fs_check_nid_range()
> > > > > > > > in f2fs_iget() to avoid getting inner inode from external interfaces.
> > > > > > >
> > > > > > > So, we don't want to check the range of inner inode numbers? What'd be the
> > > > > > > way to check it's okay?
> > > > > >
> > > > > > For node_ino, meta_ino, root_ino, we have checked them in sanity_check_raw_super()
> > > > > > as below:
> > > > > >
> > > > > > 	/* check reserved ino info */
> > > > > > 	if (le32_to_cpu(raw_super->node_ino) != 1 ||
> > > > > > 		le32_to_cpu(raw_super->meta_ino) != 2 ||
> > > > > > 		le32_to_cpu(raw_super->root_ino) != 3) {
> > > > > > 		f2fs_info(sbi, "Invalid Fs Meta Ino: node(%u) meta(%u) root(%u)",
> > > > > > 			  le32_to_cpu(raw_super->node_ino),
> > > > > > 			  le32_to_cpu(raw_super->meta_ino),
> > > > > > 			  le32_to_cpu(raw_super->root_ino));
> > > > > > 		return -EFSCORRUPTED;
> > > > > > 	}
> > > > > >
> > > > > > compressed_ino should always be NM_I(sbi)->max_nid, it can be checked in
> > > > > > f2fs_init_compress_inode()?
> > > > >
> > > > > Hmm, I'm not sure whether we really need this patch, since it'd look better
> > > > > to handle all the iget with single f2fs_iget?
> > > >
> > > > Well, the main concern is previously f2fs_iget() won't check validation for inner
> > > > inode due to it will skip do_read_inode() - f2fs_check_nid_range(), so that in a
> > > > fuzzed image, caller may pass inner ino into f2fs_iget(), result in incorrect use
> > > > of inner inode. So I add f2fs_check_nid_range() in prior to f2fs_iget_inner() in
> > > > f2fs_iget() as below to detect and avoid this case.
> > >
> > > FWIW, sanity_check_raw_super() checked the inode numbers.
> >
> > However, previously, f2fs_iget() will return inner inode to caller directly, if caller
> > passes meta_ino, node_ino or compress_ino to f2fs_iget()?
>
> Do you want to do sanity check on corrupted dentry? If so, how about checking

Yes, including:

- corrupted ino in dentry
- corrupted ino of orphan inode

As I replied in this thread:

https://lore.kernel.org/lkml/b1c74dc1-8d01-9041-9469-036273c5618d@xxxxxxxxxx/

> it in f2fs_iget instead?
>
> 	if (is_meta_ino(ino)) {
> 		if (!(inode->i_state & I_NEW)
> 			return -EFSCORRUPTED;
> 		goto make_now;
> 	}

Fine to me, let me revise in v3.

Thanks,

> > Thanks,
> >
> > > > > > > > +struct inode *f2fs_iget(struct super_block *sb, unsigned long ino)
> > > > > > > > +{
> > > > > > > > +	int ret;
> > > > > > > > +
> > > > > > > > +	ret = f2fs_check_nid_range(F2FS_SB(sb), ino);
> > > > > > > > +	if (ret)
> > > > > > > > +		return ERR_PTR(ret);
> > > > > > > > +
> > > > > > > > +	return f2fs_iget_inner(sb, ino);