[PATCH] RDMA/cma: Rollback source IP address if failing to acquire device

[PATCH] RDMA/cma: Rollback source IP address if failing to acquire device

[Myungho Jung]

If cma_acquire_dev_by_src_ip() returns error in addr_handler(), the
device state changes back to RDMA_CM_ADDR_BOUND but the resolved source
IP address is still left. After that, if rdma_destroy_id() is called
after rdma_listen(), the device is freed without removed from
listen_any_list in cma_cancel_operation(). Revert to the previous IP
address if acquiring device fails.

Reported-by: syzbot+f3ce716af730c8f96637@syzkaller.appspotmail.com
Signed-off-by: Myungho Jung <mhjungk@gmail.com>
---
 drivers/infiniband/core/cma.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
index 63a7cc00bae0..d27c3b154e71 100644
--- a/drivers/infiniband/core/cma.c
+++ b/drivers/infiniband/core/cma.c
@@ -2963,13 +2963,17 @@ static void addr_handler(int status, struct sockaddr *src_addr,
 {
 	struct rdma_id_private *id_priv = context;
 	struct rdma_cm_event event = {};
+	struct sockaddr *addr;
+	struct sockaddr_storage old_addr;
 
 	mutex_lock(&id_priv->handler_mutex);
 	if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
 			   RDMA_CM_ADDR_RESOLVED))
 		goto out;
 
-	memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
+	addr = cma_src_addr(id_priv);
+	memcpy(&old_addr, addr, rdma_addr_size(addr));
+	memcpy(addr, src_addr, rdma_addr_size(src_addr));
 	if (!status && !id_priv->cma_dev) {
 		status = cma_acquire_dev_by_src_ip(id_priv);
 		if (status)
@@ -2980,6 +2984,8 @@ static void addr_handler(int status, struct sockaddr *src_addr,
 	}
 
 	if (status) {
+		memcpy(addr, &old_addr,
+		       rdma_addr_size((struct sockaddr *)&old_addr));
 		if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
 				   RDMA_CM_ADDR_BOUND))
 			goto out;
-- 
2.17.1

RE: [PATCH] RDMA/cma: Rollback source IP address if failing to acquire device

[Parav Pandit]

> -----Original Message-----
> From: linux-rdma-owner@vger.kernel.org <linux-rdma-
> owner@vger.kernel.org> On Behalf Of Myungho Jung
> Sent: Friday, January 4, 2019 12:46 AM
> To: Doug Ledford <dledford@redhat.com>; Jason Gunthorpe <jgg@ziepe.ca>
> Cc: linux-rdma@vger.kernel.org; linux-kernel@vger.kernel.org
> Subject: [PATCH] RDMA/cma: Rollback source IP address if failing to acquire
> device
> 
> If cma_acquire_dev_by_src_ip() returns error in addr_handler(), the device
> state changes back to RDMA_CM_ADDR_BOUND but the resolved source IP
> address is still left. After that, if rdma_destroy_id() is called after
> rdma_listen(), the device is freed without removed from listen_any_list in
> cma_cancel_operation(). Revert to the previous IP address if acquiring device
> fails.
> 
> Reported-by: syzbot+f3ce716af730c8f96637@syzkaller.appspotmail.com
> Signed-off-by: Myungho Jung <mhjungk@gmail.com>
> ---
>  drivers/infiniband/core/cma.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
> index 63a7cc00bae0..d27c3b154e71 100644
> --- a/drivers/infiniband/core/cma.c
> +++ b/drivers/infiniband/core/cma.c
> @@ -2963,13 +2963,17 @@ static void addr_handler(int status, struct
> sockaddr *src_addr,  {
>  	struct rdma_id_private *id_priv = context;
>  	struct rdma_cm_event event = {};
> +	struct sockaddr *addr;
> +	struct sockaddr_storage old_addr;
> 
>  	mutex_lock(&id_priv->handler_mutex);
>  	if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
>  			   RDMA_CM_ADDR_RESOLVED))
>  		goto out;
> 
> -	memcpy(cma_src_addr(id_priv), src_addr,
> rdma_addr_size(src_addr));
> +	addr = cma_src_addr(id_priv);
> +	memcpy(&old_addr, addr, rdma_addr_size(addr));
Please add a comment here in the patch, why we need to store the old src address and restore back.
/*
  * Store the previous src address, so that if we fail to acquire matching rdma device,
  * old address can be restored back, which helps to cancel the cma listen operation
  * correctly.
  */
> +	memcpy(addr, src_addr, rdma_addr_size(src_addr));
>  	if (!status && !id_priv->cma_dev) {
>  		status = cma_acquire_dev_by_src_ip(id_priv);
>  		if (status)
> @@ -2980,6 +2984,8 @@ static void addr_handler(int status, struct sockaddr
> *src_addr,
>  	}
> 
>  	if (status) {
> +		memcpy(addr, &old_addr,
> +		       rdma_addr_size((struct sockaddr *)&old_addr));
>  		if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
>  				   RDMA_CM_ADDR_BOUND))
>  			goto out;
> --
> 2.17.1

Reviewed-by: Parav Pandit <parav@mellanox.com>

Re: [PATCH] RDMA/cma: Rollback source IP address if failing to acquire device

[Myungho Jung]

On Wed, Jan 09, 2019 at 01:28:34PM +0000, Parav Pandit wrote:
> 
> 
> > -----Original Message-----
> > From: linux-rdma-owner@vger.kernel.org <linux-rdma-
> > owner@vger.kernel.org> On Behalf Of Myungho Jung
> > Sent: Friday, January 4, 2019 12:46 AM
> > To: Doug Ledford <dledford@redhat.com>; Jason Gunthorpe <jgg@ziepe.ca>
> > Cc: linux-rdma@vger.kernel.org; linux-kernel@vger.kernel.org
> > Subject: [PATCH] RDMA/cma: Rollback source IP address if failing to acquire
> > device
> > 
> > If cma_acquire_dev_by_src_ip() returns error in addr_handler(), the device
> > state changes back to RDMA_CM_ADDR_BOUND but the resolved source IP
> > address is still left. After that, if rdma_destroy_id() is called after
> > rdma_listen(), the device is freed without removed from listen_any_list in
> > cma_cancel_operation(). Revert to the previous IP address if acquiring device
> > fails.
> > 
> > Reported-by: syzbot+f3ce716af730c8f96637@syzkaller.appspotmail.com
> > Signed-off-by: Myungho Jung <mhjungk@gmail.com>
> > ---
> >  drivers/infiniband/core/cma.c | 8 +++++++-
> >  1 file changed, 7 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
> > index 63a7cc00bae0..d27c3b154e71 100644
> > --- a/drivers/infiniband/core/cma.c
> > +++ b/drivers/infiniband/core/cma.c
> > @@ -2963,13 +2963,17 @@ static void addr_handler(int status, struct
> > sockaddr *src_addr,  {
> >  	struct rdma_id_private *id_priv = context;
> >  	struct rdma_cm_event event = {};
> > +	struct sockaddr *addr;
> > +	struct sockaddr_storage old_addr;
> > 
> >  	mutex_lock(&id_priv->handler_mutex);
> >  	if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
> >  			   RDMA_CM_ADDR_RESOLVED))
> >  		goto out;
> > 
> > -	memcpy(cma_src_addr(id_priv), src_addr,
> > rdma_addr_size(src_addr));
> > +	addr = cma_src_addr(id_priv);
> > +	memcpy(&old_addr, addr, rdma_addr_size(addr));
> Please add a comment here in the patch, why we need to store the old src address and restore back.
> /*
>   * Store the previous src address, so that if we fail to acquire matching rdma device,
>   * old address can be restored back, which helps to cancel the cma listen operation
>   * correctly.
>   */
> > +	memcpy(addr, src_addr, rdma_addr_size(src_addr));
> >  	if (!status && !id_priv->cma_dev) {
> >  		status = cma_acquire_dev_by_src_ip(id_priv);
> >  		if (status)
> > @@ -2980,6 +2984,8 @@ static void addr_handler(int status, struct sockaddr
> > *src_addr,
> >  	}
> > 
> >  	if (status) {
> > +		memcpy(addr, &old_addr,
> > +		       rdma_addr_size((struct sockaddr *)&old_addr));
> >  		if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
> >  				   RDMA_CM_ADDR_BOUND))
> >  			goto out;
> > --
> > 2.17.1
> 
> Reviewed-by: Parav Pandit <parav@mellanox.com>

Hi Parav,

Thanks for review. Let me update and resubmit patch.

Thanks,
Myungho