Re: [PATCH] nfsd: fix memory corruption due to uninitialized variable

[Junxiao Bi <junxiao.bi@oracle.com>]

On 01/20/2015 08:23 PM, Jeff Layton wrote:
> On Tue, 20 Jan 2015 19:49:47 +0800
> Junxiao Bi <junxiao.bi@oracle.com> wrote:
>
>> On 01/19/2015 10:29 PM, Jeff Layton wrote:
>>> On Mon, 19 Jan 2015 09:17:51 +0800
>>> Junxiao Bi <junxiao.bi@oracle.com> wrote:
>>>
>>>> On 01/18/2015 10:43 PM, Trond Myklebust wrote:
>>>>> On Sun, Jan 18, 2015 at 7:29 AM, Junxiao Bi <junxiao.bi@oracle.com> wrote:
>>>>>> nfsd4_decode_open() doesn't initialize variable open->op_file and
>>>>>> open->op_stp, they are initialized in nfsd4_process_open1(), but if
>>>>>> any error happens before initializing them, nfsd4_open() will call
>>>>>> into nfsd4_cleanup_open_state() and corrupt the memory.
>>>>>>
>>>>>> Since nfsd4_process_open1() will initialize these two variables and
>>>>>> open->op_openowner, make them default to null at the beginning.
>>>>>>
>>>>>> Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com>
>>>>>> ---
>>>>>>    fs/nfsd/nfs4state.c |    4 ++++
>>>>>>    1 file changed, 4 insertions(+)
>>>>>>
>>>>>> diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
>>>>>> index c06a1ba..6e74a91 100644
>>>>>> --- a/fs/nfsd/nfs4state.c
>>>>>> +++ b/fs/nfsd/nfs4state.c
>>>>>> @@ -3547,6 +3547,10 @@ nfsd4_process_open1(struct nfsd4_compound_state *cstate,
>>>>>>           struct nfs4_openowner *oo = NULL;
>>>>>>           __be32 status;
>>>>>>
>>>>>> +       open->op_file = NULL;
>>>>>> +       open->op_openowner = NULL;
>>>>>> +       open->op_stp = NULL;
>>>>>> +
>>>>>>           if (STALE_CLIENTID(&open->op_clientid, nn))
>>>>>>                   return nfserr_stale_clientid;
>>>>>>           /*
>>>>> Have you ever seen an instance of this corruption? I would have
>>>>> thought that the kzalloc() in nfsd4_decode_compound() and/or the
>>>>> earlier memset() in svc_process_common() would ensure that these
>>>>> fields are always initialised to NULL.
>>>> Yes, we got the following panic from 3.8.13. The bad pointer
>>>> open->op_stp was freed into kmem_cache array_cache, and was allocated to
>>>> next "op_stp" allocation request which triggered the panic.
>>>>
>>>>
>>>> @ PID: 21663  TASK: ffff8809fe6103c0  CPU: 0   COMMAND: "nfsd"
>>>> @ #0 [ffff8809fe613980] machine_kexec at ffffffff810421d9
>>>> @ #1 [ffff8809fe6139f0] crash_kexec at ffffffff810c9d39
>>>> @ #2 [ffff8809fe613ac0] oops_end at ffffffff81599298
>>>> @ #3 [ffff8809fe613af0] die at ffffffff8101870b
>>>> @ #4 [ffff8809fe613b20] do_general_protection at ffffffff8159906c
>>>> @ #5 [ffff8809fe613b50] general_protection at ffffffff81598668
>>>> @    [exception RIP: init_stid+14]
>>>> @    RIP: ffffffffa058247e  RSP: ffff8809fe613c08  RFLAGS: 00010292
>>>> @    RAX: 0000000000000000  RBX: 736e61727465722c  RCX: 0000000000000000
>>>> @    RDX: 0000000000000001  RSI: ffff8808e433a800  RDI: 736e61727465722c
>>>> @    RBP: ffff8809fe613c28   R8: ffff880a01469000   R9: 0000000000000000
>>>> @    R10: 0000000000000000  R11: 0000000000000000  R12: ffff8808e19821a0
>>>> @    R13: ffff8809aa40f3a8  R14: ffff8809fd781040  R15: ffff8809aafc9c98
>>>> @    ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
>>>> @ #6 [ffff8809fe613c30] nfsd4_process_open2 at ffffffffa0588123 [nfsd]
>>>> @ #7 [ffff8809fe613d00] nfsd4_open at ffffffffa0577e82 [nfsd]
>>>> @ #8 [ffff8809fe613d50] nfsd4_proc_compound at ffffffffa0575de8 [nfsd]
>>>> @ #9 [ffff8809fe613db0] nfsd_dispatch at ffffffffa056429b [nfsd]
>>>> @ #10 [ffff8809fe613df0] svc_process_common at ffffffffa04afd14 [sunrpc]
>>>> @ #11 [ffff8809fe613e70] svc_process at ffffffffa04b034f [sunrpc]
>>>> @ #12 [ffff8809fe613e90] nfsd at ffffffffa05649ff [nfsd]
>>>> @ #13 [ffff8809fe613ec0] kthread at ffffffff81082f4e
>>>> @ #14 [ffff8809fe613f50] ret_from_fork at ffffffff815a09ac
>>>>
>>>> Thanks,
>>>> Junxiao.
>>>>
>>>>> Cheers
>>>>>     Trond
>>>>>
>>> I agree with Trond. This patch doesn't make much sense.
>>>
>>> Why isn't that memset in svc_process_common() zeroing this out? If this
>>> is a bug in the open codepath, then it's almost certainly a bug for
>>> other compound ops. I'd suggest doing a bit more investigative work and
>>> see if you can figure out why that isn't working as expected...
>> Found the cause, this issue should have been fix by the following
>> commit. This fix is not merged in 3.8.13. Thanks for you and Trond
>> review it.
>>
>> commit 5d6031ca742f9f07b9c9d9322538619f3bd155ac
>> Author: J. Bruce Fields <bfields@redhat.com>
>> Date:   Thu Jul 17 16:20:39 2014 -0400
>>
>>       nfsd4: zero op arguments beyond the 8th compound op
>>
>>       The first 8 ops of the compound are zeroed since they're a part of the
>>       argument that's zeroed by the
>>
>>           memset(rqstp->rq_argp, 0, procp->pc_argsize);
>>
>>       in svc_process_common().  But we handle larger compounds by allocating
>>       the memory on the fly in nfsd4_decode_compound().  Other than code
>>       recently fixed by 01529e3f8179 "NFSD: Fix memory leak in encoding
>> denied
>>       lock", I don't know of any examples of code depending on this
>>       initialization. But it definitely seems possible, and I'd rather be
>>       safe.
>>
>>       Compounds this long are unusual so I'm much more worried about failure
>>       in this poorly tested cases than about an insignificant performance
>> hit.
>>
>>       Signed-off-by: J. Bruce Fields <bfields@redhat.com>
>>
>> diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
>> index 01023a5..628b430 100644
>> --- a/fs/nfsd/nfs4xdr.c
>> +++ b/fs/nfsd/nfs4xdr.c
>> @@ -1635,7 +1635,7 @@ nfsd4_decode_compound(struct nfsd4_compoundargs *argp)
>>                   goto xdr_error;
>>
>>           if (argp->opcnt > ARRAY_SIZE(argp->iops)) {
>> -               argp->ops = kmalloc(argp->opcnt * sizeof(*argp->ops),
>> GFP_KERNEL);
>> +               argp->ops = kzalloc(argp->opcnt * sizeof(*argp->ops),
>> GFP_KERNEL);
>>                   if (!argp->ops) {
>>                           argp->ops = argp->iops;
>>                           dprintk("nfsd: couldn't allocate room for
>> COMPOUND\n");
>>
>> Thanks,
>> Junxiao.
> Yes, that patch looks fine, and I'm pretty sure it'd be ok for stable.
yes.
> I don't think v3.8 is being maintained anymore though, is it?
Used by us internal.

Thanks,
Junxiao.
>