[PATCH] NFS: hard-code init_net for NFS callback transports

[PATCH] NFS: hard-code init_net for NFS callback transports

[Stanislav Kinsbursky]

This patch set is a bug fix for Bruce's 3.5 branch.
I'll rebase this for 3.4 and resend.

In case of destroying mount namespace on child reaper exit, nsproxy is zeroed
to the point already. So, dereferencing of it is invalid.
This patch hard-code "init_net" for all network namespace references for NFS
callback services. This will be fixed with proper NFS callback
containerization.

Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
---
 fs/nfs/callback.c |   11 +++++------
 1 files changed, 5 insertions(+), 6 deletions(-)

diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c
index 970659d..23ff18f 100644
--- a/fs/nfs/callback.c
+++ b/fs/nfs/callback.c
@@ -17,7 +17,6 @@
 #include <linux/kthread.h>
 #include <linux/sunrpc/svcauth_gss.h>
 #include <linux/sunrpc/bc_xprt.h>
-#include <linux/nsproxy.h>
 
 #include <net/inet_sock.h>
 
@@ -107,7 +106,7 @@ nfs4_callback_up(struct svc_serv *serv, struct rpc_xprt *xprt)
 {
 	int ret;
 
-	ret = svc_create_xprt(serv, "tcp", xprt->xprt_net, PF_INET,
+	ret = svc_create_xprt(serv, "tcp", &init_net, PF_INET,
 				nfs_callback_set_tcpport, SVC_SOCK_ANONYMOUS);
 	if (ret <= 0)
 		goto out_err;
@@ -115,7 +114,7 @@ nfs4_callback_up(struct svc_serv *serv, struct rpc_xprt *xprt)
 	dprintk("NFS: Callback listener port = %u (af %u)\n",
 			nfs_callback_tcpport, PF_INET);
 
-	ret = svc_create_xprt(serv, "tcp", xprt->xprt_net, PF_INET6,
+	ret = svc_create_xprt(serv, "tcp", &init_net, PF_INET6,
 				nfs_callback_set_tcpport, SVC_SOCK_ANONYMOUS);
 	if (ret > 0) {
 		nfs_callback_tcpport6 = ret;
@@ -184,7 +183,7 @@ nfs41_callback_up(struct svc_serv *serv, struct rpc_xprt *xprt)
 	 * fore channel connection.
 	 * Returns the input port (0) and sets the svc_serv bc_xprt on success
 	 */
-	ret = svc_create_xprt(serv, "tcp-bc", xprt->xprt_net, PF_INET, 0,
+	ret = svc_create_xprt(serv, "tcp-bc", &init_net, PF_INET, 0,
 			      SVC_SOCK_ANONYMOUS);
 	if (ret < 0) {
 		rqstp = ERR_PTR(ret);
@@ -254,7 +253,7 @@ int nfs_callback_up(u32 minorversion, struct rpc_xprt *xprt)
 	char svc_name[12];
 	int ret = 0;
 	int minorversion_setup;
-	struct net *net = current->nsproxy->net_ns;
+	struct net *net = &init_net;
 
 	mutex_lock(&nfs_callback_mutex);
 	if (cb_info->users++ || cb_info->task != NULL) {
@@ -330,7 +329,7 @@ void nfs_callback_down(int minorversion)
 	cb_info->users--;
 	if (cb_info->users == 0 && cb_info->task != NULL) {
 		kthread_stop(cb_info->task);
-		svc_shutdown_net(cb_info->serv, current->nsproxy->net_ns);
+		svc_shutdown_net(cb_info->serv, &init_net);
 		svc_exit_thread(cb_info->rqst);
 		cb_info->serv = NULL;
 		cb_info->rqst = NULL;

Re: [PATCH] NFS: hard-code init_net for NFS callback transports

[J. Bruce Fields]

On Fri, Jun 01, 2012 at 01:55:47PM +0400, Stanislav Kinsbursky wrote:
> This patch set is a bug fix for Bruce's 3.5 branch.
> I'll rebase this for 3.4 and resend.

Apologies for neglecting this, I'm done with the bakeathon and should be
back to this soon.

But, I'm a bit confused: you say you'll rebase this, but I've only seen
two rebased patches for 3.4, and neither looks like a simple rebase of
this one.

OK, I'll look more closely and probably I'll figure out.  But in
general, a little more detail in changelogs would help me process
patches more quickly:

	- Please make sure it's clear whether a given patch is a bug
	  fix, a new feature, or code reshuffling that isn't intended to
	  change behavior.  If it's a bug fix, tell me whether we've had
	  the bug forever or whether it's a recent regression.  And
	  (relevant here) if it's a recent regression, reference the ID
	  of the commit that introduced it.
	- If something is a backport, please reference in the changelog
	  the commit ID of the backported patch, especially if it's not
	  obviously the same.

--b.
	

> 
> In case of destroying mount namespace on child reaper exit, nsproxy is zeroed
> to the point already. So, dereferencing of it is invalid.
> This patch hard-code "init_net" for all network namespace references for NFS
> callback services. This will be fixed with proper NFS callback
> containerization.
> 
> Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
> ---
>  fs/nfs/callback.c |   11 +++++------
>  1 files changed, 5 insertions(+), 6 deletions(-)
> 
> diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c
> index 970659d..23ff18f 100644
> --- a/fs/nfs/callback.c
> +++ b/fs/nfs/callback.c
> @@ -17,7 +17,6 @@
>  #include <linux/kthread.h>
>  #include <linux/sunrpc/svcauth_gss.h>
>  #include <linux/sunrpc/bc_xprt.h>
> -#include <linux/nsproxy.h>
>  
>  #include <net/inet_sock.h>
>  
> @@ -107,7 +106,7 @@ nfs4_callback_up(struct svc_serv *serv, struct rpc_xprt *xprt)
>  {
>  	int ret;
>  
> -	ret = svc_create_xprt(serv, "tcp", xprt->xprt_net, PF_INET,
> +	ret = svc_create_xprt(serv, "tcp", &init_net, PF_INET,
>  				nfs_callback_set_tcpport, SVC_SOCK_ANONYMOUS);
>  	if (ret <= 0)
>  		goto out_err;
> @@ -115,7 +114,7 @@ nfs4_callback_up(struct svc_serv *serv, struct rpc_xprt *xprt)
>  	dprintk("NFS: Callback listener port = %u (af %u)\n",
>  			nfs_callback_tcpport, PF_INET);
>  
> -	ret = svc_create_xprt(serv, "tcp", xprt->xprt_net, PF_INET6,
> +	ret = svc_create_xprt(serv, "tcp", &init_net, PF_INET6,
>  				nfs_callback_set_tcpport, SVC_SOCK_ANONYMOUS);
>  	if (ret > 0) {
>  		nfs_callback_tcpport6 = ret;
> @@ -184,7 +183,7 @@ nfs41_callback_up(struct svc_serv *serv, struct rpc_xprt *xprt)
>  	 * fore channel connection.
>  	 * Returns the input port (0) and sets the svc_serv bc_xprt on success
>  	 */
> -	ret = svc_create_xprt(serv, "tcp-bc", xprt->xprt_net, PF_INET, 0,
> +	ret = svc_create_xprt(serv, "tcp-bc", &init_net, PF_INET, 0,
>  			      SVC_SOCK_ANONYMOUS);
>  	if (ret < 0) {
>  		rqstp = ERR_PTR(ret);
> @@ -254,7 +253,7 @@ int nfs_callback_up(u32 minorversion, struct rpc_xprt *xprt)
>  	char svc_name[12];
>  	int ret = 0;
>  	int minorversion_setup;
> -	struct net *net = current->nsproxy->net_ns;
> +	struct net *net = &init_net;
>  
>  	mutex_lock(&nfs_callback_mutex);
>  	if (cb_info->users++ || cb_info->task != NULL) {
> @@ -330,7 +329,7 @@ void nfs_callback_down(int minorversion)
>  	cb_info->users--;
>  	if (cb_info->users == 0 && cb_info->task != NULL) {
>  		kthread_stop(cb_info->task);
> -		svc_shutdown_net(cb_info->serv, current->nsproxy->net_ns);
> +		svc_shutdown_net(cb_info->serv, &init_net);
>  		svc_exit_thread(cb_info->rqst);
>  		cb_info->serv = NULL;
>  		cb_info->rqst = NULL;
> 

Re: [PATCH] NFS: hard-code init_net for NFS callback transports

[J. Bruce Fields]

On Mon, Jun 11, 2012 at 08:37:18AM -0400, J. Bruce Fields wrote:
> On Fri, Jun 01, 2012 at 01:55:47PM +0400, Stanislav Kinsbursky wrote:
> > This patch set is a bug fix for Bruce's 3.5 branch.
> > I'll rebase this for 3.4 and resend.
> 
> Apologies for neglecting this, I'm done with the bakeathon and should be
> back to this soon.
> 
> But, I'm a bit confused: you say you'll rebase this, but I've only seen
> two rebased patches for 3.4, and neither looks like a simple rebase of
> this one.

Oh, I see; the backport of "SUNRPC: new svc_bind() routine introduced"
is a combination of the patch by the same name upstream with this one.

In future for backports I'd rather keep a one-to-one correspondance
between the original patches and the backported patches.  (Even if that
means 4.2 might not be as "bisectable".)

Also, get this patch into stable by just adding a "cc:
stable@vger.kernel.org" with the signed-off at the end of this patch.

I've gone ahead and fixed that up; results in the "for-3.4" and
"for-3.5" branches of

	git://linux-nfs.org/~bfields/linux-topics.git

Let me know if anything looks wrong.

--b.

> 
> OK, I'll look more closely and probably I'll figure out.  But in
> general, a little more detail in changelogs would help me process
> patches more quickly:
> 
> 	- Please make sure it's clear whether a given patch is a bug
> 	  fix, a new feature, or code reshuffling that isn't intended to
> 	  change behavior.  If it's a bug fix, tell me whether we've had
> 	  the bug forever or whether it's a recent regression.  And
> 	  (relevant here) if it's a recent regression, reference the ID
> 	  of the commit that introduced it.
> 	- If something is a backport, please reference in the changelog
> 	  the commit ID of the backported patch, especially if it's not
> 	  obviously the same.
> 
> --b.
> 	
> 
> > 
> > In case of destroying mount namespace on child reaper exit, nsproxy is zeroed
> > to the point already. So, dereferencing of it is invalid.
> > This patch hard-code "init_net" for all network namespace references for NFS
> > callback services. This will be fixed with proper NFS callback
> > containerization.
> > 
> > Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
> > ---
> >  fs/nfs/callback.c |   11 +++++------
> >  1 files changed, 5 insertions(+), 6 deletions(-)
> > 
> > diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c
> > index 970659d..23ff18f 100644
> > --- a/fs/nfs/callback.c
> > +++ b/fs/nfs/callback.c
> > @@ -17,7 +17,6 @@
> >  #include <linux/kthread.h>
> >  #include <linux/sunrpc/svcauth_gss.h>
> >  #include <linux/sunrpc/bc_xprt.h>
> > -#include <linux/nsproxy.h>
> >  
> >  #include <net/inet_sock.h>
> >  
> > @@ -107,7 +106,7 @@ nfs4_callback_up(struct svc_serv *serv, struct rpc_xprt *xprt)
> >  {
> >  	int ret;
> >  
> > -	ret = svc_create_xprt(serv, "tcp", xprt->xprt_net, PF_INET,
> > +	ret = svc_create_xprt(serv, "tcp", &init_net, PF_INET,
> >  				nfs_callback_set_tcpport, SVC_SOCK_ANONYMOUS);
> >  	if (ret <= 0)
> >  		goto out_err;
> > @@ -115,7 +114,7 @@ nfs4_callback_up(struct svc_serv *serv, struct rpc_xprt *xprt)
> >  	dprintk("NFS: Callback listener port = %u (af %u)\n",
> >  			nfs_callback_tcpport, PF_INET);
> >  
> > -	ret = svc_create_xprt(serv, "tcp", xprt->xprt_net, PF_INET6,
> > +	ret = svc_create_xprt(serv, "tcp", &init_net, PF_INET6,
> >  				nfs_callback_set_tcpport, SVC_SOCK_ANONYMOUS);
> >  	if (ret > 0) {
> >  		nfs_callback_tcpport6 = ret;
> > @@ -184,7 +183,7 @@ nfs41_callback_up(struct svc_serv *serv, struct rpc_xprt *xprt)
> >  	 * fore channel connection.
> >  	 * Returns the input port (0) and sets the svc_serv bc_xprt on success
> >  	 */
> > -	ret = svc_create_xprt(serv, "tcp-bc", xprt->xprt_net, PF_INET, 0,
> > +	ret = svc_create_xprt(serv, "tcp-bc", &init_net, PF_INET, 0,
> >  			      SVC_SOCK_ANONYMOUS);
> >  	if (ret < 0) {
> >  		rqstp = ERR_PTR(ret);
> > @@ -254,7 +253,7 @@ int nfs_callback_up(u32 minorversion, struct rpc_xprt *xprt)
> >  	char svc_name[12];
> >  	int ret = 0;
> >  	int minorversion_setup;
> > -	struct net *net = current->nsproxy->net_ns;
> > +	struct net *net = &init_net;
> >  
> >  	mutex_lock(&nfs_callback_mutex);
> >  	if (cb_info->users++ || cb_info->task != NULL) {
> > @@ -330,7 +329,7 @@ void nfs_callback_down(int minorversion)
> >  	cb_info->users--;
> >  	if (cb_info->users == 0 && cb_info->task != NULL) {
> >  		kthread_stop(cb_info->task);
> > -		svc_shutdown_net(cb_info->serv, current->nsproxy->net_ns);
> > +		svc_shutdown_net(cb_info->serv, &init_net);
> >  		svc_exit_thread(cb_info->rqst);
> >  		cb_info->serv = NULL;
> >  		cb_info->rqst = NULL;
> > 

Re: [PATCH] NFS: hard-code init_net for NFS callback transports

[Stanislav Kinsbursky]

12.06.2012 19:53, J. Bruce Fields написал:
> On Mon, Jun 11, 2012 at 08:37:18AM -0400, J. Bruce Fields wrote:
>> On Fri, Jun 01, 2012 at 01:55:47PM +0400, Stanislav Kinsbursky wrote:
>>> This patch set is a bug fix for Bruce's 3.5 branch.
>>> I'll rebase this for 3.4 and resend.
>>
>> Apologies for neglecting this, I'm done with the bakeathon and should be
>> back to this soon.
>>
>> But, I'm a bit confused: you say you'll rebase this, but I've only seen
>> two rebased patches for 3.4, and neither looks like a simple rebase of
>> this one.
>
> Oh, I see; the backport of "SUNRPC: new svc_bind() routine introduced"
> is a combination of the patch by the same name upstream with this one.
>
> In future for backports I'd rather keep a one-to-one correspondance
> between the original patches and the backported patches.  (Even if that
> means 4.2 might not be as "bisectable".)
>
> Also, get this patch into stable by just adding a "cc:
> stable@vger.kernel.org" with the signed-off at the end of this patch.
>
> I've gone ahead and fixed that up; results in the "for-3.4" and
> "for-3.5" branches of
>
> 	git://linux-nfs.org/~bfields/linux-topics.git
>
> Let me know if anything looks wrong.
>


Hi, Bruce. Thanks for you time and guidance with this patch set.
But I'm a bit confused - what I have to do now? Take the patches from your 3.4 
branch and send them to "stable"?


> --b.
>
>>
>> OK, I'll look more closely and probably I'll figure out.  But in
>> general, a little more detail in changelogs would help me process
>> patches more quickly:
>>
>> 	- Please make sure it's clear whether a given patch is a bug
>> 	  fix, a new feature, or code reshuffling that isn't intended to
>> 	  change behavior.  If it's a bug fix, tell me whether we've had
>> 	  the bug forever or whether it's a recent regression.  And
>> 	  (relevant here) if it's a recent regression, reference the ID
>> 	  of the commit that introduced it.
>> 	- If something is a backport, please reference in the changelog
>> 	  the commit ID of the backported patch, especially if it's not
>> 	  obviously the same.
>>
>> --b.
>> 	
>>
>>>
>>> In case of destroying mount namespace on child reaper exit, nsproxy is zeroed
>>> to the point already. So, dereferencing of it is invalid.
>>> This patch hard-code "init_net" for all network namespace references for NFS
>>> callback services. This will be fixed with proper NFS callback
>>> containerization.
>>>
>>> Signed-off-by: Stanislav Kinsbursky<skinsbursky@parallels.com>
>>> ---
>>>   fs/nfs/callback.c |   11 +++++------
>>>   1 files changed, 5 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c
>>> index 970659d..23ff18f 100644
>>> --- a/fs/nfs/callback.c
>>> +++ b/fs/nfs/callback.c
>>> @@ -17,7 +17,6 @@
>>>   #include<linux/kthread.h>
>>>   #include<linux/sunrpc/svcauth_gss.h>
>>>   #include<linux/sunrpc/bc_xprt.h>
>>> -#include<linux/nsproxy.h>
>>>
>>>   #include<net/inet_sock.h>
>>>
>>> @@ -107,7 +106,7 @@ nfs4_callback_up(struct svc_serv *serv, struct rpc_xprt *xprt)
>>>   {
>>>   	int ret;
>>>
>>> -	ret = svc_create_xprt(serv, "tcp", xprt->xprt_net, PF_INET,
>>> +	ret = svc_create_xprt(serv, "tcp",&init_net, PF_INET,
>>>   				nfs_callback_set_tcpport, SVC_SOCK_ANONYMOUS);
>>>   	if (ret<= 0)
>>>   		goto out_err;
>>> @@ -115,7 +114,7 @@ nfs4_callback_up(struct svc_serv *serv, struct rpc_xprt *xprt)
>>>   	dprintk("NFS: Callback listener port = %u (af %u)\n",
>>>   			nfs_callback_tcpport, PF_INET);
>>>
>>> -	ret = svc_create_xprt(serv, "tcp", xprt->xprt_net, PF_INET6,
>>> +	ret = svc_create_xprt(serv, "tcp",&init_net, PF_INET6,
>>>   				nfs_callback_set_tcpport, SVC_SOCK_ANONYMOUS);
>>>   	if (ret>  0) {
>>>   		nfs_callback_tcpport6 = ret;
>>> @@ -184,7 +183,7 @@ nfs41_callback_up(struct svc_serv *serv, struct rpc_xprt *xprt)
>>>   	 * fore channel connection.
>>>   	 * Returns the input port (0) and sets the svc_serv bc_xprt on success
>>>   	 */
>>> -	ret = svc_create_xprt(serv, "tcp-bc", xprt->xprt_net, PF_INET, 0,
>>> +	ret = svc_create_xprt(serv, "tcp-bc",&init_net, PF_INET, 0,
>>>   			      SVC_SOCK_ANONYMOUS);
>>>   	if (ret<  0) {
>>>   		rqstp = ERR_PTR(ret);
>>> @@ -254,7 +253,7 @@ int nfs_callback_up(u32 minorversion, struct rpc_xprt *xprt)
>>>   	char svc_name[12];
>>>   	int ret = 0;
>>>   	int minorversion_setup;
>>> -	struct net *net = current->nsproxy->net_ns;
>>> +	struct net *net =&init_net;
>>>
>>>   	mutex_lock(&nfs_callback_mutex);
>>>   	if (cb_info->users++ || cb_info->task != NULL) {
>>> @@ -330,7 +329,7 @@ void nfs_callback_down(int minorversion)
>>>   	cb_info->users--;
>>>   	if (cb_info->users == 0&&  cb_info->task != NULL) {
>>>   		kthread_stop(cb_info->task);
>>> -		svc_shutdown_net(cb_info->serv, current->nsproxy->net_ns);
>>> +		svc_shutdown_net(cb_info->serv,&init_net);
>>>   		svc_exit_thread(cb_info->rqst);
>>>   		cb_info->serv = NULL;
>>>   		cb_info->rqst = NULL;
>>>


-- 
Best regards,
Stanislav Kinsbursky

Re: [PATCH] NFS: hard-code init_net for NFS callback transports

[J. Bruce Fields]

On Wed, Jun 13, 2012 at 12:16:07PM +0400, Stanislav Kinsbursky wrote:
> 12.06.2012 19:53, J. Bruce Fields написал:
> >On Mon, Jun 11, 2012 at 08:37:18AM -0400, J. Bruce Fields wrote:
> >>On Fri, Jun 01, 2012 at 01:55:47PM +0400, Stanislav Kinsbursky wrote:
> >>>This patch set is a bug fix for Bruce's 3.5 branch.
> >>>I'll rebase this for 3.4 and resend.
> >>
> >>Apologies for neglecting this, I'm done with the bakeathon and should be
> >>back to this soon.
> >>
> >>But, I'm a bit confused: you say you'll rebase this, but I've only seen
> >>two rebased patches for 3.4, and neither looks like a simple rebase of
> >>this one.
> >
> >Oh, I see; the backport of "SUNRPC: new svc_bind() routine introduced"
> >is a combination of the patch by the same name upstream with this one.
> >
> >In future for backports I'd rather keep a one-to-one correspondance
> >between the original patches and the backported patches.  (Even if that
> >means 4.2 might not be as "bisectable".)
> >
> >Also, get this patch into stable by just adding a "cc:
> >stable@vger.kernel.org" with the signed-off at the end of this patch.
> >
> >I've gone ahead and fixed that up; results in the "for-3.4" and
> >"for-3.5" branches of
> >
> >	git://linux-nfs.org/~bfields/linux-topics.git
> >
> >Let me know if anything looks wrong.
> >
> 
> 
> Hi, Bruce. Thanks for you time and guidance with this patch set.
> But I'm a bit confused - what I have to do now? Take the patches
> from your 3.4 branch and send them to "stable"?

No, I'll take care of them.--b.

--b.