diff mbox

[RFC] nfsd: fix possible dereference of static NULL nfsd_serv pointer

Message ID 20120706134548.7307.1133.stgit@localhost.localdomain (mailing list archive)
State New, archived
Headers show

Commit Message

Stanislav Kinsbursky July 6, 2012, 1:45 p.m. UTC 
This is a bug fix for 3.5 kernel.
In case on NFSd service start failure svc_shutdown_net() will call svc_destroy
callback and zeroize global nfsd_serv pointer, this in turn will lead to Oops
in svc_destroy().

This patch is marked as RFC, because to many lines were changed. It can be
easely simplified if requested.
Moreover, NFSd service shutdown is going to be converted into csomething on
per-net basis.

Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
---
 fs/nfsd/nfssvc.c |   14 ++++++++------
 1 files changed, 8 insertions(+), 6 deletions(-)


--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

J. Bruce Fields July 6, 2012, 5:26 p.m. UTC | #1 
On Fri, Jul 06, 2012 at 05:45:56PM +0400, Stanislav Kinsbursky wrote:
> This is a bug fix for 3.5 kernel.
> In case on NFSd service start failure svc_shutdown_net() will call svc_destroy
> callback and zeroize global nfsd_serv pointer, this in turn will lead to Oops
> in svc_destroy().
> 
> This patch is marked as RFC, because to many lines were changed. It can be
> easely simplified if requested.
> Moreover, NFSd service shutdown is going to be converted into csomething on
> per-net basis.

Doesn't this leave error paths in e.g. __write_ports_addfd() and
__write_ports_addxprt() unfixed?

I'm inclined to just submit your original fix (split up as in the last
version I sent) for 3.5 if you don't object.

--b.

> 
> Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
> ---
>  fs/nfsd/nfssvc.c |   14 ++++++++------
>  1 files changed, 8 insertions(+), 6 deletions(-)
> 
> diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> index ee709fc..526a4aa 100644
> --- a/fs/nfsd/nfssvc.c
> +++ b/fs/nfsd/nfssvc.c
> @@ -446,6 +446,7 @@ nfsd_svc(unsigned short port, int nrservs)
>  	int	error;
>  	bool	nfsd_up_before;
>  	struct net *net = &init_net;
> +	struct svc_serv *serv = nfsd_serv;
>  
>  	mutex_lock(&nfsd_mutex);
>  	dprintk("nfsd: creating service\n");
> @@ -454,7 +455,7 @@ nfsd_svc(unsigned short port, int nrservs)
>  	if (nrservs > NFSD_MAXSERVS)
>  		nrservs = NFSD_MAXSERVS;
>  	error = 0;
> -	if (nrservs == 0 && nfsd_serv == NULL)
> +	if (nrservs == 0 && serv == NULL)
>  		goto out;
>  
>  	error = nfsd_create_serv();
> @@ -464,23 +465,24 @@ nfsd_svc(unsigned short port, int nrservs)
>  	nfsd_up_before = nfsd_up;
>  
>  	error = nfsd_startup(port, nrservs);
> +	error = -EINVAL;
>  	if (error)
>  		goto out_destroy;
> -	error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
> +	error = svc_set_num_threads(serv, NULL, nrservs);
>  	if (error)
>  		goto out_shutdown;
>  	/* We are holding a reference to nfsd_serv which
>  	 * we don't want to count in the return value,
>  	 * so subtract 1
>  	 */
> -	error = nfsd_serv->sv_nrthreads - 1;
> +	error = serv->sv_nrthreads - 1;
>  out_shutdown:
>  	if (error < 0 && !nfsd_up_before)
>  		nfsd_shutdown();
>  out_destroy:
> -	if (nfsd_serv->sv_nrthreads == 1)
> -		svc_shutdown_net(nfsd_serv, net);
> -	svc_destroy(nfsd_serv);		/* Release server */
> +	if (serv->sv_nrthreads == 1)
> +		svc_shutdown_net(serv, net);
> +	svc_destroy(serv);		/* Release server */
>  out:
>  	mutex_unlock(&nfsd_mutex);
>  	return error;
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Stanislav Kinsbursky July 7, 2012, 5:27 a.m. UTC | #2 
06.07.2012 21:26, J. Bruce Fields ?????:
> On Fri, Jul 06, 2012 at 05:45:56PM +0400, Stanislav Kinsbursky wrote:
>> This is a bug fix for 3.5 kernel.
>> In case on NFSd service start failure svc_shutdown_net() will call svc_destroy
>> callback and zeroize global nfsd_serv pointer, this in turn will lead to Oops
>> in svc_destroy().
>>
>> This patch is marked as RFC, because to many lines were changed. It can be
>> easely simplified if requested.
>> Moreover, NFSd service shutdown is going to be converted into csomething on
>> per-net basis.
> Doesn't this leave error paths in e.g. __write_ports_addfd() and
> __write_ports_addxprt() unfixed?

Yes, sure it does...

> I'm inclined to just submit your original fix (split up as in the last
> version I sent) for 3.5 if you don't object.

Not at all.
Thanks, Bruce.

> --b.
>
>> Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
>> ---
>>   fs/nfsd/nfssvc.c |   14 ++++++++------
>>   1 files changed, 8 insertions(+), 6 deletions(-)
>>
>> diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
>> index ee709fc..526a4aa 100644
>> --- a/fs/nfsd/nfssvc.c
>> +++ b/fs/nfsd/nfssvc.c
>> @@ -446,6 +446,7 @@ nfsd_svc(unsigned short port, int nrservs)
>>   	int	error;
>>   	bool	nfsd_up_before;
>>   	struct net *net = &init_net;
>> +	struct svc_serv *serv = nfsd_serv;
>>   
>>   	mutex_lock(&nfsd_mutex);
>>   	dprintk("nfsd: creating service\n");
>> @@ -454,7 +455,7 @@ nfsd_svc(unsigned short port, int nrservs)
>>   	if (nrservs > NFSD_MAXSERVS)
>>   		nrservs = NFSD_MAXSERVS;
>>   	error = 0;
>> -	if (nrservs == 0 && nfsd_serv == NULL)
>> +	if (nrservs == 0 && serv == NULL)
>>   		goto out;
>>   
>>   	error = nfsd_create_serv();
>> @@ -464,23 +465,24 @@ nfsd_svc(unsigned short port, int nrservs)
>>   	nfsd_up_before = nfsd_up;
>>   
>>   	error = nfsd_startup(port, nrservs);
>> +	error = -EINVAL;
>>   	if (error)
>>   		goto out_destroy;
>> -	error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
>> +	error = svc_set_num_threads(serv, NULL, nrservs);
>>   	if (error)
>>   		goto out_shutdown;
>>   	/* We are holding a reference to nfsd_serv which
>>   	 * we don't want to count in the return value,
>>   	 * so subtract 1
>>   	 */
>> -	error = nfsd_serv->sv_nrthreads - 1;
>> +	error = serv->sv_nrthreads - 1;
>>   out_shutdown:
>>   	if (error < 0 && !nfsd_up_before)
>>   		nfsd_shutdown();
>>   out_destroy:
>> -	if (nfsd_serv->sv_nrthreads == 1)
>> -		svc_shutdown_net(nfsd_serv, net);
>> -	svc_destroy(nfsd_serv);		/* Release server */
>> +	if (serv->sv_nrthreads == 1)
>> +		svc_shutdown_net(serv, net);
>> +	svc_destroy(serv);		/* Release server */
>>   out:
>>   	mutex_unlock(&nfsd_mutex);
>>   	return error;
>>


--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
J. Bruce Fields July 10, 2012, 3:59 p.m. UTC | #3 
On Sat, Jul 07, 2012 at 09:27:30AM +0400, Stanislav Kinsbursky wrote:
> 06.07.2012 21:26, J. Bruce Fields ?????:
> >On Fri, Jul 06, 2012 at 05:45:56PM +0400, Stanislav Kinsbursky wrote:
> >>This is a bug fix for 3.5 kernel.
> >>In case on NFSd service start failure svc_shutdown_net() will call svc_destroy
> >>callback and zeroize global nfsd_serv pointer, this in turn will lead to Oops
> >>in svc_destroy().
> >>
> >>This patch is marked as RFC, because to many lines were changed. It can be
> >>easely simplified if requested.
> >>Moreover, NFSd service shutdown is going to be converted into csomething on
> >>per-net basis.
> >Doesn't this leave error paths in e.g. __write_ports_addfd() and
> >__write_ports_addxprt() unfixed?
> 
> Yes, sure it does...
> 
> >I'm inclined to just submit your original fix (split up as in the last
> >version I sent) for 3.5 if you don't object.
> 
> Not at all.
> Thanks, Bruce.

OK.  Actually, Linus is making noise about a release in the next week or
two, so given that this is about error paths, I'm going to queue it up
for the 3.6 and cc it to stable.  It'll end up in 3.5.x pretty quickly
that way anyway.

--b.
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Stanislav Kinsbursky July 10, 2012, 4:04 p.m. UTC | #4 
10.07.2012 19:59, J. Bruce Fields ?????:
> On Sat, Jul 07, 2012 at 09:27:30AM +0400, Stanislav Kinsbursky wrote:
>> 06.07.2012 21:26, J. Bruce Fields ?????:
>>> On Fri, Jul 06, 2012 at 05:45:56PM +0400, Stanislav Kinsbursky wrote:
>>>> This is a bug fix for 3.5 kernel.
>>>> In case on NFSd service start failure svc_shutdown_net() will call svc_destroy
>>>> callback and zeroize global nfsd_serv pointer, this in turn will lead to Oops
>>>> in svc_destroy().
>>>>
>>>> This patch is marked as RFC, because to many lines were changed. It can be
>>>> easely simplified if requested.
>>>> Moreover, NFSd service shutdown is going to be converted into csomething on
>>>> per-net basis.
>>> Doesn't this leave error paths in e.g. __write_ports_addfd() and
>>> __write_ports_addxprt() unfixed?
>>
>> Yes, sure it does...
>>
>>> I'm inclined to just submit your original fix (split up as in the last
>>> version I sent) for 3.5 if you don't object.
>>
>> Not at all.
>> Thanks, Bruce.
>
> OK.  Actually, Linus is making noise about a release in the next week or
> two, so given that this is about error paths, I'm going to queue it up
> for the 3.6 and cc it to stable.  It'll end up in 3.5.x pretty quickly
> that way anyway.
>

Ok, Bruce. Sounds good.
diff mbox

Patch

diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
index ee709fc..526a4aa 100644
--- a/fs/nfsd/nfssvc.c
+++ b/fs/nfsd/nfssvc.c
@@ -446,6 +446,7 @@  nfsd_svc(unsigned short port, int nrservs)
 	int	error;
 	bool	nfsd_up_before;
 	struct net *net = &init_net;
+	struct svc_serv *serv = nfsd_serv;
 
 	mutex_lock(&nfsd_mutex);
 	dprintk("nfsd: creating service\n");
@@ -454,7 +455,7 @@  nfsd_svc(unsigned short port, int nrservs)
 	if (nrservs > NFSD_MAXSERVS)
 		nrservs = NFSD_MAXSERVS;
 	error = 0;
-	if (nrservs == 0 && nfsd_serv == NULL)
+	if (nrservs == 0 && serv == NULL)
 		goto out;
 
 	error = nfsd_create_serv();
@@ -464,23 +465,24 @@  nfsd_svc(unsigned short port, int nrservs)
 	nfsd_up_before = nfsd_up;
 
 	error = nfsd_startup(port, nrservs);
+	error = -EINVAL;
 	if (error)
 		goto out_destroy;
-	error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
+	error = svc_set_num_threads(serv, NULL, nrservs);
 	if (error)
 		goto out_shutdown;
 	/* We are holding a reference to nfsd_serv which
 	 * we don't want to count in the return value,
 	 * so subtract 1
 	 */
-	error = nfsd_serv->sv_nrthreads - 1;
+	error = serv->sv_nrthreads - 1;
 out_shutdown:
 	if (error < 0 && !nfsd_up_before)
 		nfsd_shutdown();
 out_destroy:
-	if (nfsd_serv->sv_nrthreads == 1)
-		svc_shutdown_net(nfsd_serv, net);
-	svc_destroy(nfsd_serv);		/* Release server */
+	if (serv->sv_nrthreads == 1)
+		svc_shutdown_net(serv, net);
+	svc_destroy(serv);		/* Release server */
 out:
 	mutex_unlock(&nfsd_mutex);
 	return error;