linux-kernel - Re: [PATCH] radix-tree: fix radix_tree_iter

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:	Tue, 19 Jul 2016 12:07:37 +0300
From:	Konstantin Khlebnikov <koct9i@...il.com>
To:	Ross Zwisler <ross.zwisler@...ux.intel.com>
Cc:	Andrey Ryabinin <aryabinin@...tuozzo.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Jan Kara <jack@...e.cz>,
	"Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>,
	"linux-mm@...ck.org" <linux-mm@...ck.org>,
	Greg Thelen <gthelen@...gle.com>,
	Suleiman Souhlal <suleiman@...gle.com>,
	syzkaller <syzkaller@...glegroups.com>,
	Kostya Serebryany <kcc@...gle.com>,
	Alexander Potapenko <glider@...gle.com>,
	Sasha Levin <sasha.levin@...cle.com>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
	Matthew Wilcox <willy@...ux.intel.com>,
	Hugh Dickins <hughd@...gle.com>,
	Stable <stable@...r.kernel.org>
Subject: Re: [PATCH] radix-tree: fix radix_tree_iter_retry() for tagged iterators.

On Tue, Jul 19, 2016 at 7:11 AM, Ross Zwisler
<ross.zwisler@...ux.intel.com> wrote:
> On Sat, Jul 16, 2016 at 04:45:31PM +0300, Konstantin Khlebnikov wrote:
>> On Fri, Jul 15, 2016 at 10:00 PM, Ross Zwisler
>> <ross.zwisler@...ux.intel.com> wrote:
> <>
>> > 3) radix_tree_iter_next() via via a non-tagged iteration like
>> > radix_tree_for_each_slot().  This currently happens in shmem_tag_pins()
>> > and shmem_partial_swap_usage().
>> >
>> > I think that this case is currently unhandled.  Unlike with
>> > radix_tree_iter_retry() case (#1 above) we can't rely on 'count' in the else
>> > case of radix_tree_next_slot() to be zero, so I think it's possible we'll end
>> > up executing code in the while() loop in radix_tree_next_slot() that assumes
>> > 'slot' is valid.
>> >
>> > I haven't actually seen this crash on a test setup, but I don't think the
>> > current code is safe.
>>
>> This is becase distance between ->index and ->next_index now could be
>> more that one?
>>
>> We could fix that by adding "iter->index = iter->next_index - 1;" into
>> radix_tree_iter_next()
>> right after updating next_index and tweak multi-order itreration logic
>> if it depends on that.
>>
>> I'd like to keep radix_tree_next_slot() as small as possible because
>> this is supposed to be a fast-path.
>
> I think it'll be exactly one?
>
>         iter->next_index = __radix_tree_iter_add(iter, 1);
>
> So iter->index will be X, iter->next_index will be X+1, accounting for the
> iterator's shift.  So, basically, whatever your height is, you'll be set up to
> process one more entry, I think.
>
> This means that radix_tree_chunk_size() will return 1.  I guess with the
> current logic this is safe:
>
> static __always_inline void **
> radix_tree_next_slot(void **slot, struct radix_tree_iter *iter, unsigned flags)
> {
> ...
>         } else {
>                 long count = radix_tree_chunk_size(iter);
>                 void *canon = slot;
>
>                 while (--count > 0) {
>                         /* code that assumes 'slot' is non-NULL */
>
> So 'count' will be 1, the prefix decrement will make it 0, so we won't execute
> the code in the while() loop.  So maybe all the cases are covered after all.
>
> It seems like we need some unit tests in tools/testing/radix-tree around this
> - I'll try and find time to add them this week.
>
> I just feel like this isn't very organized.  We have a lot of code in
> radix_tree_next_slot() that assumes that 'slot' is non-NULL, but we don't
> check it anywhere.  We know it *can* be NULL, but we just happen to have
> things set up so that none of the code that uses 'slot' is executed.
>
> I personally feel like a quick check for slot==NULL at the beginning of the
> function is the simplest way to keep ourselves safe, and it doesn't seem like
> we'd be adding that much overhead.

Either fix is fine now. I working on better design for multiorder iterator which
moves all that logic from radix_tree_next_slot() into radix_tree_next_chunk().

Most likely I'll change tree structure a little. For example I think sibling
entries chould hold offset to head entry and order rather than a pointer to it.
Or maybe size: support of non-power-of-2 entries is interesting feature too.