linux-kernel - Re: [PATCH v3] drivers: net: prevent tun_get

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]

Message-ID: <64c7d788.190a0220.3c2cf.5d7f@mx.google.com>
Date:   Mon, 31 Jul 2023 18:47:16 +0300
From:   Andrew Kanner <andrew.kanner@...il.com>
To:     David Ahern <dsahern@...il.com>
Cc:     Jesper Dangaard Brouer <jbrouer@...hat.com>,
        Paolo Abeni <pabeni@...hat.com>,
        Jason Wang <jasowang@...hat.com>, brouer@...hat.com,
        davem@...emloft.net, edumazet@...gle.com, kuba@...nel.org,
        netdev@...r.kernel.org, linux-kernel@...r.kernel.org,
        linux-kernel-mentees@...ts.linuxfoundation.org,
        syzbot+f817490f5bd20541b90a@...kaller.appspotmail.com,
        John Fastabend <john.fastabend@...il.com>
Subject: Re: [PATCH v3] drivers: net: prevent tun_get_user() to exceed xdp
 size limits

On Thu, Jul 27, 2023 at 06:11:57PM -0600, David Ahern wrote:
> On 7/27/23 5:48 PM, Andrew Kanner wrote:
> > 
> > Thanks, everyone.
> > 
> > If we summarize the discussion - there are 3 issues here:
> > 1. tun_can_build_skb() doesn't count XDP_PACKET_HEADROOM (minor and
> >    most trivial)
> > 2. WARN_ON_ONCE from net/core/filter.c, which may be too strict / not
> >    needed at all.
> > 3. strange behaviour with reallocationg SKB (65007 -> 131072)
> 
> I believe that happens because of the current skb size and the need to
> expand it to account for the XDP headroom makes the allocation go over
> 64kB. Since tun is given the packet via a write call there are no header
> markers to allocate separate space for headers and data (e.g. like TCP
> does with 32kB data segments).


Yes, this is exactly what you suspected. In pskb_expand_head() ->
kmalloc_reserve() I have these values initially:
(gdb) p *size
$13 = 65408
(gdb) p obj_size
$16 = 65728

and it will do:
    data = kmalloc_reserve(&size, gfp_mask, NUMA_NO_NODE, NULL);
...
	obj_size = SKB_HEAD_ALIGN(*size);
...
	*size = obj_size = kmalloc_size_roundup(obj_size);

(gdb) p *size
$22 = 131072

So this is kmalloc_size_roundup() doing this math with the following:
   /* Above the smaller buckets, size is a multiple of page size. */                                                                                                                           │
   if (size > KMALLOC_MAX_CACHE_SIZE)                                                                                                                                                          │
      return PAGE_SIZE << get_order(size);

> > 
> > I can check these issues. I have to dive a little deeper with 2-3,
> > most likely with kgdb and syzkaller repro. But seems this is not
> > somewhat urgent and lives quite a long time without being noticed.
> > 
> > BTW: Attached the ftrace logs using the original syzkaller repro
> > (starting with tun_get_user()). They answer Jesper's question about
> > contiguous physical memory allocation (kmem_cache_alloc_node() /
> > kmalloc_reserve()). But I'll check it one more time before submitting
> > a new PATCH V4 or another patch / patch series.
> > 
> 

I see no other bugs in math, so not sure wether it should be fixed. Is
it ok and expected to roundup the memory allocation?


-- 
Andrew Kanner