Re: [PATCH] hfs: fix array out of bounds read of array extent

[Al Viro]

On Wed, Oct 17, 2018 at 03:01:17PM -0700, Andrew Morton wrote:
> On Fri, 31 Aug 2018 15:05:38 +0100 Colin King <colin.king@xxxxxxxxxxxxx> wrote:
> 
> > From: Colin Ian King <colin.king@xxxxxxxxxxxxx>
> > 
> > Currently extent and index i are both being incremented causing
> > an array out of bounds read on extent[i]. Fix this by removing
> > the extraneous increment of extent.
> > 
> > Detected by CoverityScan, CID#711541 ("Out of bounds read")
> > 
> > Fixes: d1081202f1d0 ("HFS rewrite")
> 
> No such commit here.  I assume this is 7cb74be6fd827e314f8.
> 
> > --- a/fs/hfs/extent.c
> > +++ b/fs/hfs/extent.c
> > @@ -300,7 +300,7 @@ int hfs_free_fork(struct super_block *sb, struct hfs_cat_file *file, int type)
> >  		return 0;
> >  
> >  	blocks = 0;
> > -	for (i = 0; i < 3; extent++, i++)
> > +	for (i = 0; i < 3; i++)
> >  		blocks += be16_to_cpu(extent[i].count);
> >  
> >  	res = hfs_free_extents(sb, extent, blocks, blocks);
> 
> Well, that's quite the bug.  Question is, why didn't anyone notice it. 
> What are the runtime effects?  A disk space leak, perhaps?
> 
> I worry a bit that, given the fs was evidently working "ok", perhaps
> this error was corrected elsewhere in the code and that "fixing" this
> site will have unexpected and undesirable runtime effects.  Can someone
> help me out here?

hfs_free_extents() seems to expect the 'offset' argument to be the
sum of ->count of 1--3 starting elements of extent array.  In case of
mismatch, it returns -EIO and that's it - hfs_free_fork() will bugger
off with -EIO at that point.  If it does match, block_nr is supposed
to be in range 0..offset and blocks offset - block_nr .. offset - 1
are freed.

So at a guess, that sucker mostly ends up leaking blocks.  Said that,
it means that the rest of hfs_free_fork() has never been tested.

I'd suggest somebody to turn that
        /* panic? */
        return -EIO;
in hfs_free_extents() into
	printk(KERN_ERR "hfs_free_extents is fucked");
	return -EIO;
and see if it's triggerable.  Then check if there's a block leak in
the reproducer, whatever it is.