Re: [PATCH 2/2] maple_tree: Fix a potential memory leak, OOB access, or other unpredictable bug

[Peng Zhang]

在 2023/4/10 23:00, Liam R. Howlett 写道:
> * Peng Zhang <perlyzhang@xxxxxxxxx> [230410 09:28]:
> > 在 2023/4/10 21:12, Liam R. Howlett 写道:
> > > * Peng Zhang <perlyzhang@xxxxxxxxx> [230410 08:58]:
> > > > 在 2023/4/10 20:43, Liam R. Howlett 写道:
> > > > > * Peng Zhang <zhangpeng.00@xxxxxxxxxxxxx> [230407 00:10]:
> > > > > > In mas_alloc_nodes(), there is such a piece of code:
> > > > > > while (requested) {
> > > > > > 	...
> > > > > > 	node->node_count = 0;
> > > > > > 	...
> > > > > > }
> > > > > You don't need to quote code in your commit message since it is
> > > > > available in the change log or in the file itself.
> > > > Ok, I will change it in the next version.
> > > > > > "node->node_count = 0" means to initialize the node_count field of the
> > > > > > new node, but the node may not be a new node. It may be a node that
> > > > > > existed before and node_count has a value, setting it to 0 will cause a
> > > > > > memory leak. At this time, mas->alloc->total will be greater than the
> > > > > > actual number of nodes in the linked list, which may cause many other
> > > > > > errors. For example, out-of-bounds access in mas_pop_node(), and
> > > > > > mas_pop_node() may return addresses that should not be used.
> > > > > > Fix it by initializing node_count only for new nodes.
> > > > > >
> > > > > > Fixes: 54a611b60590 ("Maple Tree: add new data structure")
> > > > > > Signed-off-by: Peng Zhang <zhangpeng.00@xxxxxxxxxxxxx>
> > > > > > Cc: <stable@xxxxxxxxxxxxxxx>
> > > > > > ---
> > > > > >     lib/maple_tree.c | 16 ++++------------
> > > > > >     1 file changed, 4 insertions(+), 12 deletions(-)
> > > > > >
> > > > > > diff --git a/lib/maple_tree.c b/lib/maple_tree.c
> > > > > > index 65fd861b30e1..9e25b3215803 100644
> > > > > > --- a/lib/maple_tree.c
> > > > > > +++ b/lib/maple_tree.c
> > > > > > @@ -1249,26 +1249,18 @@ static inline void mas_alloc_nodes(struct ma_state *mas, gfp_t gfp)
> > > > > >     	node = mas->alloc;
> > > > > >     	node->request_count = 0;
> > > > > >     	while (requested) {
> > > > > > -		max_req = MAPLE_ALLOC_SLOTS;
> > > > > > -		if (node->node_count) {
> > > > > > -			unsigned int offset = node->node_count;
> > > > > > -
> > > > > > -			slots = (void **)&node->slot[offset];
> > > > > > -			max_req -= offset;
> > > > > > -		} else {
> > > > > > -			slots = (void **)&node->slot;
> > > > > > -		}
> > > > > > -
> > > > > > +		max_req = MAPLE_ALLOC_SLOTS - node->node_count;
> > > > > > +		slots = (void **)&node->slot[node->node_count];
> > > > > Thanks, this is much cleaner.
> > > > >
> > > > > >     		max_req = min(requested, max_req);
> > > > > >     		count = mt_alloc_bulk(gfp, max_req, slots);
> > > > > >     		if (!count)
> > > > > >     			goto nomem_bulk;
> > > > > > +		if (node->node_count == 0)
> > > > > > +			node->slot[0]->node_count = 0;
> > > > > >     		node->node_count += count;
> > > > > >     		allocated += count;
> > > > > >     		node = node->slot[0];
> > > > > > -		node->node_count = 0;
> > > > > > -		node->request_count = 0;
> > > > > Why are we not clearing request_count anymore?
> > > > Because the node pointed to by the variable "node"
> > > > must not be the head node of the linked list at
> > > > this time, we only need to maintain the information
> > > > of the head node.
> > > Right, at this time it is not the head node, but could it become the
> > > head node with invalid data?  I think it can, because we don't
> > > explicitly set it in mas_pop_node()?
> > 1. Actually in mas_pop_node(), when a node becomes the head node,
> >     we initialize its total field and request_count field.
> Only if there is a request_count to begin with, right?
>
> > 2. The total field and request_count field of any non-head node,
> >     even if we initialize it, cannot be considered a valid value.
> >     Imagine if the request_count of the head node is changed, then
> >     we don't actually change the request_count of the non-head nodes,
> >     so it is an invalid value anyway.
> When we pop a node, we record the requested value and only initialize it
> to the recorded value + 1 if it wasn't zero.  So if there are no
> requests, we don't initialize it.

Yes, you are right.
I neglected that if request_count is equal to 0,
the request_count field of the new head node will not be set.
There are many implementation details of maple_tree,
which is quite error-prone.
I will modify it in the next version.

Thanks.

> This works because of the zeroing of that request_count that you removed
> here.  But it was, as you pointed out, not always using the right node.
> I think this needs to be moved to your new 'if' statement.
>
> > > In any case, be sure to mention that you make a change like this in the
> > > change log, like "Drop setting the resquest_count as it is unnecessary
> > > because.." in a new paragraph, so that it is not missed.
> > I thought it was a small change that wasn't written in the changelog.
> > In the next version and any future patches, I will write down the
> > details of any changes.
> >
> > Thanks.
> >
> > > > > >     		requested -= count;
> > > > > >     	}
> > > > > >     	mas->alloc->total = allocated;
> > > > > > --
> > > > > > 2.20.1