Re: [PATCH] bcachefs: Fix NULL ptr dereference in btree_node_iter_and_journal_peek

[Kent Overstreet]

On Sat, Oct 26, 2024 at 12:23:40AM +0000, Piotr Zalewski wrote:
> 
> 
> 
> 
> 
> Sent with Proton Mail secure email.
> 
> On Saturday, October 26th, 2024 at 2:16 AM, Kent Overstreet <kent.overstreet@xxxxxxxxx> wrote:
> 
> > On Fri, Oct 25, 2024 at 08:11:50PM -0400, Kent Overstreet wrote:
> > 
> > > On Wed, Oct 23, 2024 at 03:33:22PM +0800, Alan Huang wrote:
> > > 
> > > > On Oct 23, 2024, at 15:21, Piotr Zalewski pZ010001011111@xxxxxxxxx wrote:
> > > > 
> > > > > Add NULL check for key returned from bch2_btree_and_journal_iter_peek in
> > > > > btree_node_iter_and_journal_peek to avoid NULL ptr dereference in
> > > > > bch2_bkey_buf_reassemble.
> > > > 
> > > > It would be helpful if the commit message explained why k.k is null in this case
> > > 
> > > This code is only for iterating over interior btree nodes - k.k is only
> > > null when we have a bad btree topology (gaps).
> > > 
> > > Piotr, could you add a comment to that effect?
> > 
> > 
> > Actually, not just that - when this happens we should flag the
> > filesystem as having topology repairs, and possibly start topology
> > repair.
> > 
> > Calling bch2_topology_error() will do that.
> > 
> > We definitely want to log an error message, too; it should reference the
> > btree node we're iterating over and explain that it's missing child
> > nodes.
> 
> Thanks for the clarification. I will send v2 tomorrow :)

Also, make sure we're returning an error code - your patch didn't do
that, we (obviously) can't continue the btree lookup.

bch2_topology_error() will give you the error code you want; the error
code will tell recovery to rewind and run topology repair (if we're in
recovery) or else something otherwise sensible.