linux-rdma.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH rc] RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests
@ 2021-09-16 18:34 Jason Gunthorpe
  2021-09-22  8:01 ` Leon Romanovsky
  0 siblings, 1 reply; 10+ messages in thread
From: Jason Gunthorpe @ 2021-09-16 18:34 UTC (permalink / raw)
  To: Dmitry Vyukov, linux-rdma; +Cc: syzbot+dc3dfba010d7671e05f5

The FSM can run in a circle allowing rdma_resolve_ip() to be called twice
on the same id_priv. While this cannot happen without going through the
work, it violates the invariant that the same address resolution
background request cannot be active twice.

       CPU 1                                  CPU 2

rdma_resolve_addr():
  RDMA_CM_IDLE -> RDMA_CM_ADDR_QUERY
  rdma_resolve_ip(addr_handler)  #1

			 process_one_req(): for #1
                          addr_handler():
                            RDMA_CM_ADDR_QUERY -> RDMA_CM_ADDR_BOUND
                            mutex_unlock(&id_priv->handler_mutex);
                            [.. handler still running ..]

rdma_resolve_addr():
  RDMA_CM_ADDR_BOUND -> RDMA_CM_ADDR_QUERY
  rdma_resolve_ip(addr_handler)
    !! two requests are now on the req_list

rdma_destroy_id():
 destroy_id_handler_unlock():
  _destroy_id():
   cma_cancel_operation():
    rdma_addr_cancel()

                          // process_one_req() self removes it
		          spin_lock_bh(&lock);
                           cancel_delayed_work(&req->work);
	                   if (!list_empty(&req->list)) == true

      ! rdma_addr_cancel() returns after process_on_req #1 is done

   kfree(id_priv)

			 process_one_req(): for #2
                          addr_handler():
	                    mutex_lock(&id_priv->handler_mutex);
                            !! Use after free on id_priv

rdma_addr_cancel() expects there to be one req on the list and only
cancels the first one. The self-removal behavior of the work only happens
after the handler has returned. This yields a situations where the
req_list can have two reqs for the same "handle" but rdma_addr_cancel()
only cancels the first one.

The second req remains active beyond rdma_destroy_id() and will
use-after-free id_priv once it inevitably triggers.

Fix this by remembering if the id_priv has called rdma_resolve_ip() and
always cancel before calling it again. This ensures the req_list never
gets more than one item in it and doesn't cost anything in the normal flow
that never uses this strange error path.

Cc: stable@vger.kernel.org
Fixes: e51060f08a61 ("IB: IP address based RDMA connection manager")
Reported-by: syzbot+dc3dfba010d7671e05f5@syzkaller.appspotmail.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/infiniband/core/cma.c      | 17 +++++++++++++++++
 drivers/infiniband/core/cma_priv.h |  1 +
 2 files changed, 18 insertions(+)

diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
index c40791baced588..751cf5ea25f296 100644
--- a/drivers/infiniband/core/cma.c
+++ b/drivers/infiniband/core/cma.c
@@ -1776,6 +1776,14 @@ static void cma_cancel_operation(struct rdma_id_private *id_priv,
 {
 	switch (state) {
 	case RDMA_CM_ADDR_QUERY:
+		/*
+		 * We can avoid doing the rdma_addr_cancel() based on state,
+		 * only RDMA_CM_ADDR_QUERY has a work that could still execute.
+		 * Notice that the addr_handler work could still be exiting
+		 * outside this state, however due to the interaction with the
+		 * handler_mutex the work is guaranteed not to touch id_priv
+		 * during exit.
+		 */
 		rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
 		break;
 	case RDMA_CM_ROUTE_QUERY:
@@ -3413,6 +3421,15 @@ int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
 		if (dst_addr->sa_family == AF_IB) {
 			ret = cma_resolve_ib_addr(id_priv);
 		} else {
+			/* The FSM can return back to RDMA_CM_ADDR_BOUND after
+			 * rdma_resolve_ip() is called, eg through the error
+			 * path in addr_handler. If this happens the existing
+			 * request must be canceled before issuing a new one.
+			 */
+			if (id_priv->used_resolve_ip)
+				rdma_addr_cancel(&id->route.addr.dev_addr);
+			else
+				id_priv->used_resolve_ip = 1;
 			ret = rdma_resolve_ip(cma_src_addr(id_priv), dst_addr,
 					      &id->route.addr.dev_addr,
 					      timeout_ms, addr_handler,
diff --git a/drivers/infiniband/core/cma_priv.h b/drivers/infiniband/core/cma_priv.h
index 5c463da9984536..f92f101ea9818f 100644
--- a/drivers/infiniband/core/cma_priv.h
+++ b/drivers/infiniband/core/cma_priv.h
@@ -91,6 +91,7 @@ struct rdma_id_private {
 	u8			afonly;
 	u8			timeout;
 	u8			min_rnr_timer;
+	u8 used_resolve_ip;
 	enum ib_gid_type	gid_type;
 
 	/*

base-commit: ad17bbef3dd573da937816edc0ab84fed6a17fa6
-- 
2.33.0


^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH rc] RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests
  2021-09-16 18:34 [PATCH rc] RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests Jason Gunthorpe
@ 2021-09-22  8:01 ` Leon Romanovsky
  2021-09-22  9:38   ` Haakon Bugge
  2021-09-22 14:41   ` Jason Gunthorpe
  0 siblings, 2 replies; 10+ messages in thread
From: Leon Romanovsky @ 2021-09-22  8:01 UTC (permalink / raw)
  To: Jason Gunthorpe; +Cc: Dmitry Vyukov, linux-rdma, syzbot+dc3dfba010d7671e05f5

On Thu, Sep 16, 2021 at 03:34:46PM -0300, Jason Gunthorpe wrote:
> The FSM can run in a circle allowing rdma_resolve_ip() to be called twice
> on the same id_priv. While this cannot happen without going through the
> work, it violates the invariant that the same address resolution
> background request cannot be active twice.
> 
>        CPU 1                                  CPU 2
> 
> rdma_resolve_addr():
>   RDMA_CM_IDLE -> RDMA_CM_ADDR_QUERY
>   rdma_resolve_ip(addr_handler)  #1
> 
> 			 process_one_req(): for #1
>                           addr_handler():
>                             RDMA_CM_ADDR_QUERY -> RDMA_CM_ADDR_BOUND
>                             mutex_unlock(&id_priv->handler_mutex);
>                             [.. handler still running ..]
> 
> rdma_resolve_addr():
>   RDMA_CM_ADDR_BOUND -> RDMA_CM_ADDR_QUERY
>   rdma_resolve_ip(addr_handler)
>     !! two requests are now on the req_list
> 
> rdma_destroy_id():
>  destroy_id_handler_unlock():
>   _destroy_id():
>    cma_cancel_operation():
>     rdma_addr_cancel()
> 
>                           // process_one_req() self removes it
> 		          spin_lock_bh(&lock);
>                            cancel_delayed_work(&req->work);
> 	                   if (!list_empty(&req->list)) == true
> 
>       ! rdma_addr_cancel() returns after process_on_req #1 is done
> 
>    kfree(id_priv)
> 
> 			 process_one_req(): for #2
>                           addr_handler():
> 	                    mutex_lock(&id_priv->handler_mutex);
>                             !! Use after free on id_priv
> 
> rdma_addr_cancel() expects there to be one req on the list and only
> cancels the first one. The self-removal behavior of the work only happens
> after the handler has returned. This yields a situations where the
> req_list can have two reqs for the same "handle" but rdma_addr_cancel()
> only cancels the first one.
> 
> The second req remains active beyond rdma_destroy_id() and will
> use-after-free id_priv once it inevitably triggers.
> 
> Fix this by remembering if the id_priv has called rdma_resolve_ip() and
> always cancel before calling it again. This ensures the req_list never
> gets more than one item in it and doesn't cost anything in the normal flow
> that never uses this strange error path.
> 
> Cc: stable@vger.kernel.org
> Fixes: e51060f08a61 ("IB: IP address based RDMA connection manager")
> Reported-by: syzbot+dc3dfba010d7671e05f5@syzkaller.appspotmail.com
> Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
> ---
>  drivers/infiniband/core/cma.c      | 17 +++++++++++++++++
>  drivers/infiniband/core/cma_priv.h |  1 +
>  2 files changed, 18 insertions(+)
> 
> diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
> index c40791baced588..751cf5ea25f296 100644
> --- a/drivers/infiniband/core/cma.c
> +++ b/drivers/infiniband/core/cma.c
> @@ -1776,6 +1776,14 @@ static void cma_cancel_operation(struct rdma_id_private *id_priv,
>  {
>  	switch (state) {
>  	case RDMA_CM_ADDR_QUERY:
> +		/*
> +		 * We can avoid doing the rdma_addr_cancel() based on state,
> +		 * only RDMA_CM_ADDR_QUERY has a work that could still execute.
> +		 * Notice that the addr_handler work could still be exiting
> +		 * outside this state, however due to the interaction with the
> +		 * handler_mutex the work is guaranteed not to touch id_priv
> +		 * during exit.
> +		 */
>  		rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
>  		break;
>  	case RDMA_CM_ROUTE_QUERY:
> @@ -3413,6 +3421,15 @@ int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
>  		if (dst_addr->sa_family == AF_IB) {
>  			ret = cma_resolve_ib_addr(id_priv);
>  		} else {
> +			/* The FSM can return back to RDMA_CM_ADDR_BOUND after
> +			 * rdma_resolve_ip() is called, eg through the error
> +			 * path in addr_handler. If this happens the existing
> +			 * request must be canceled before issuing a new one.
> +			 */
> +			if (id_priv->used_resolve_ip)
> +				rdma_addr_cancel(&id->route.addr.dev_addr);
> +			else
> +				id_priv->used_resolve_ip = 1;

Why don't you never clear this field? If you assume that this is one lifetime
event, can you please add a comment with an explanation "why"?

Thanks

>  			ret = rdma_resolve_ip(cma_src_addr(id_priv), dst_addr,
>  					      &id->route.addr.dev_addr,
>  					      timeout_ms, addr_handler,
> diff --git a/drivers/infiniband/core/cma_priv.h b/drivers/infiniband/core/cma_priv.h
> index 5c463da9984536..f92f101ea9818f 100644
> --- a/drivers/infiniband/core/cma_priv.h
> +++ b/drivers/infiniband/core/cma_priv.h
> @@ -91,6 +91,7 @@ struct rdma_id_private {
>  	u8			afonly;
>  	u8			timeout;
>  	u8			min_rnr_timer;
> +	u8 used_resolve_ip;
>  	enum ib_gid_type	gid_type;
>  
>  	/*
> 
> base-commit: ad17bbef3dd573da937816edc0ab84fed6a17fa6
> -- 
> 2.33.0
> 

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH rc] RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests
  2021-09-22  8:01 ` Leon Romanovsky
@ 2021-09-22  9:38   ` Haakon Bugge
  2021-09-22 14:44     ` Jason Gunthorpe
  2021-09-22 14:41   ` Jason Gunthorpe
  1 sibling, 1 reply; 10+ messages in thread
From: Haakon Bugge @ 2021-09-22  9:38 UTC (permalink / raw)
  To: Leon Romanovsky
  Cc: Jason Gunthorpe, Dmitry Vyukov, OFED mailing list,
	syzbot+dc3dfba010d7671e05f5



> On 22 Sep 2021, at 10:01, Leon Romanovsky <leon@kernel.org> wrote:
> 
> On Thu, Sep 16, 2021 at 03:34:46PM -0300, Jason Gunthorpe wrote:
>> The FSM can run in a circle allowing rdma_resolve_ip() to be called twice
>> on the same id_priv. While this cannot happen without going through the
>> work, it violates the invariant that the same address resolution
>> background request cannot be active twice.
>> 
>>       CPU 1                                  CPU 2
>> 
>> rdma_resolve_addr():
>>  RDMA_CM_IDLE -> RDMA_CM_ADDR_QUERY
>>  rdma_resolve_ip(addr_handler)  #1
>> 
>> 			 process_one_req(): for #1
>>                          addr_handler():
>>                            RDMA_CM_ADDR_QUERY -> RDMA_CM_ADDR_BOUND
>>                            mutex_unlock(&id_priv->handler_mutex);
>>                            [.. handler still running ..]
>> 
>> rdma_resolve_addr():
>>  RDMA_CM_ADDR_BOUND -> RDMA_CM_ADDR_QUERY
>>  rdma_resolve_ip(addr_handler)
>>    !! two requests are now on the req_list
>> 
>> rdma_destroy_id():
>> destroy_id_handler_unlock():
>>  _destroy_id():
>>   cma_cancel_operation():
>>    rdma_addr_cancel()
>> 
>>                          // process_one_req() self removes it
>> 		          spin_lock_bh(&lock);
>>                           cancel_delayed_work(&req->work);
>> 	                   if (!list_empty(&req->list)) == true
>> 
>>      ! rdma_addr_cancel() returns after process_on_req #1 is done
>> 
>>   kfree(id_priv)
>> 
>> 			 process_one_req(): for #2
>>                          addr_handler():
>> 	                    mutex_lock(&id_priv->handler_mutex);
>>                            !! Use after free on id_priv
>> 
>> rdma_addr_cancel() expects there to be one req on the list and only
>> cancels the first one. The self-removal behavior of the work only happens
>> after the handler has returned. This yields a situations where the
>> req_list can have two reqs for the same "handle" but rdma_addr_cancel()
>> only cancels the first one.
>> 
>> The second req remains active beyond rdma_destroy_id() and will
>> use-after-free id_priv once it inevitably triggers.
>> 
>> Fix this by remembering if the id_priv has called rdma_resolve_ip() and
>> always cancel before calling it again. This ensures the req_list never
>> gets more than one item in it and doesn't cost anything in the normal flow
>> that never uses this strange error path.
>> 
>> Cc: stable@vger.kernel.org
>> Fixes: e51060f08a61 ("IB: IP address based RDMA connection manager")
>> Reported-by: syzbot+dc3dfba010d7671e05f5@syzkaller.appspotmail.com
>> Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
>> ---
>> drivers/infiniband/core/cma.c      | 17 +++++++++++++++++
>> drivers/infiniband/core/cma_priv.h |  1 +
>> 2 files changed, 18 insertions(+)
>> 
>> diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
>> index c40791baced588..751cf5ea25f296 100644
>> --- a/drivers/infiniband/core/cma.c
>> +++ b/drivers/infiniband/core/cma.c
>> @@ -1776,6 +1776,14 @@ static void cma_cancel_operation(struct rdma_id_private *id_priv,
>> {
>> 	switch (state) {
>> 	case RDMA_CM_ADDR_QUERY:
>> +		/*
>> +		 * We can avoid doing the rdma_addr_cancel() based on state,
>> +		 * only RDMA_CM_ADDR_QUERY has a work that could still execute.
>> +		 * Notice that the addr_handler work could still be exiting
>> +		 * outside this state, however due to the interaction with the
>> +		 * handler_mutex the work is guaranteed not to touch id_priv
>> +		 * during exit.
>> +		 */
>> 		rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
>> 		break;
>> 	case RDMA_CM_ROUTE_QUERY:
>> @@ -3413,6 +3421,15 @@ int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
>> 		if (dst_addr->sa_family == AF_IB) {
>> 			ret = cma_resolve_ib_addr(id_priv);
>> 		} else {
>> +			/* The FSM can return back to RDMA_CM_ADDR_BOUND after
>> +			 * rdma_resolve_ip() is called, eg through the error
>> +			 * path in addr_handler. If this happens the existing
>> +			 * request must be canceled before issuing a new one.
>> +			 */
>> +			if (id_priv->used_resolve_ip)
>> +				rdma_addr_cancel(&id->route.addr.dev_addr);
>> +			else
>> +				id_priv->used_resolve_ip = 1;
> 
> Why don't you never clear this field? If you assume that this is one lifetime
> event, can you please add a comment with an explanation "why"?

Adding to that, don't you need {READ,WRITE}_ONCE when accessing used_resolve_ip? Or will the write to it obtain global visibility because mutex_unlock(&ctx->mutex) is executed before any other context can read it?


Thxs, Håkon

> 
> Thanks
> 
>> 			ret = rdma_resolve_ip(cma_src_addr(id_priv), dst_addr,
>> 					      &id->route.addr.dev_addr,
>> 					      timeout_ms, addr_handler,
>> diff --git a/drivers/infiniband/core/cma_priv.h b/drivers/infiniband/core/cma_priv.h
>> index 5c463da9984536..f92f101ea9818f 100644
>> --- a/drivers/infiniband/core/cma_priv.h
>> +++ b/drivers/infiniband/core/cma_priv.h
>> @@ -91,6 +91,7 @@ struct rdma_id_private {
>> 	u8			afonly;
>> 	u8			timeout;
>> 	u8			min_rnr_timer;
>> +	u8 used_resolve_ip;
>> 	enum ib_gid_type	gid_type;
>> 
>> 	/*
>> 
>> base-commit: ad17bbef3dd573da937816edc0ab84fed6a17fa6
>> -- 
>> 2.33.0


^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH rc] RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests
  2021-09-22  8:01 ` Leon Romanovsky
  2021-09-22  9:38   ` Haakon Bugge
@ 2021-09-22 14:41   ` Jason Gunthorpe
  2021-09-23  5:49     ` Leon Romanovsky
  1 sibling, 1 reply; 10+ messages in thread
From: Jason Gunthorpe @ 2021-09-22 14:41 UTC (permalink / raw)
  To: Leon Romanovsky; +Cc: Dmitry Vyukov, linux-rdma, syzbot+dc3dfba010d7671e05f5

On Wed, Sep 22, 2021 at 11:01:39AM +0300, Leon Romanovsky wrote:

> > +			/* The FSM can return back to RDMA_CM_ADDR_BOUND after
> > +			 * rdma_resolve_ip() is called, eg through the error
> > +			 * path in addr_handler. If this happens the existing
> > +			 * request must be canceled before issuing a new one.
> > +			 */
> > +			if (id_priv->used_resolve_ip)
> > +				rdma_addr_cancel(&id->route.addr.dev_addr);
> > +			else
> > +				id_priv->used_resolve_ip = 1;
> 
> Why don't you never clear this field?

The only case where it can be cleared is if we have called
rdma_addr_cancel(), and since this is the only place that does it and
immediately calls rdma_resolve_ip() again, there is no reason to ever
clear it.

Jason

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH rc] RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests
  2021-09-22  9:38   ` Haakon Bugge
@ 2021-09-22 14:44     ` Jason Gunthorpe
  0 siblings, 0 replies; 10+ messages in thread
From: Jason Gunthorpe @ 2021-09-22 14:44 UTC (permalink / raw)
  To: Haakon Bugge
  Cc: Leon Romanovsky, Dmitry Vyukov, OFED mailing list,
	syzbot+dc3dfba010d7671e05f5

On Wed, Sep 22, 2021 at 09:38:40AM +0000, Haakon Bugge wrote:
> >> @@ -3413,6 +3421,15 @@ int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
> >> 		if (dst_addr->sa_family == AF_IB) {
> >> 			ret = cma_resolve_ib_addr(id_priv);
> >> 		} else {
> >> +			/* The FSM can return back to RDMA_CM_ADDR_BOUND after
> >> +			 * rdma_resolve_ip() is called, eg through the error
> >> +			 * path in addr_handler. If this happens the existing
> >> +			 * request must be canceled before issuing a new one.
> >> +			 */
> >> +			if (id_priv->used_resolve_ip)
> >> +				rdma_addr_cancel(&id->route.addr.dev_addr);
> >> +			else
> >> +				id_priv->used_resolve_ip = 1;
> > 
> > Why don't you never clear this field? If you assume that this is one lifetime
> > event, can you please add a comment with an explanation "why"?
> 
> Adding to that, don't you need {READ,WRITE}_ONCE when accessing
> used_resolve_ip? 

The FSM logic guarentees there is no concurrent access here, this is
the only thread that can be in this state at this point.

> Or will the write to it obtain global visibility because
> mutex_unlock(&ctx->mutex) is executed before any other context can
> read it?

Global visibility flows indirectly through the rdma_resolve_ip() to
the work. Basically when the rdma_resolve_ip schedules the work it
does a full release, then the work does a spinlock/unlock which is
another full release, finally the next time we go through this
function it does another spinlock/unlock which will act as ancquire
for this store.

Jason

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH rc] RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests
  2021-09-22 14:41   ` Jason Gunthorpe
@ 2021-09-23  5:49     ` Leon Romanovsky
  2021-09-23 11:45       ` Jason Gunthorpe
  0 siblings, 1 reply; 10+ messages in thread
From: Leon Romanovsky @ 2021-09-23  5:49 UTC (permalink / raw)
  To: Jason Gunthorpe; +Cc: Dmitry Vyukov, linux-rdma, syzbot+dc3dfba010d7671e05f5

On Wed, Sep 22, 2021 at 11:41:19AM -0300, Jason Gunthorpe wrote:
> On Wed, Sep 22, 2021 at 11:01:39AM +0300, Leon Romanovsky wrote:
> 
> > > +			/* The FSM can return back to RDMA_CM_ADDR_BOUND after
> > > +			 * rdma_resolve_ip() is called, eg through the error
> > > +			 * path in addr_handler. If this happens the existing
> > > +			 * request must be canceled before issuing a new one.
> > > +			 */
> > > +			if (id_priv->used_resolve_ip)
> > > +				rdma_addr_cancel(&id->route.addr.dev_addr);
> > > +			else
> > > +				id_priv->used_resolve_ip = 1;
> > 
> > Why don't you never clear this field?
> 
> The only case where it can be cleared is if we have called
> rdma_addr_cancel(), and since this is the only place that does it and
> immediately calls rdma_resolve_ip() again, there is no reason to ever
> clear it.

IMHO, it is better to clear instead to rely on "the only place" semantic.

Thanks

> 
> Jason

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH rc] RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests
  2021-09-23  5:49     ` Leon Romanovsky
@ 2021-09-23 11:45       ` Jason Gunthorpe
  2021-09-23 18:15         ` Leon Romanovsky
  0 siblings, 1 reply; 10+ messages in thread
From: Jason Gunthorpe @ 2021-09-23 11:45 UTC (permalink / raw)
  To: Leon Romanovsky; +Cc: Dmitry Vyukov, linux-rdma, syzbot+dc3dfba010d7671e05f5

On Thu, Sep 23, 2021 at 08:49:06AM +0300, Leon Romanovsky wrote:
> On Wed, Sep 22, 2021 at 11:41:19AM -0300, Jason Gunthorpe wrote:
> > On Wed, Sep 22, 2021 at 11:01:39AM +0300, Leon Romanovsky wrote:
> > 
> > > > +			/* The FSM can return back to RDMA_CM_ADDR_BOUND after
> > > > +			 * rdma_resolve_ip() is called, eg through the error
> > > > +			 * path in addr_handler. If this happens the existing
> > > > +			 * request must be canceled before issuing a new one.
> > > > +			 */
> > > > +			if (id_priv->used_resolve_ip)
> > > > +				rdma_addr_cancel(&id->route.addr.dev_addr);
> > > > +			else
> > > > +				id_priv->used_resolve_ip = 1;
> > > 
> > > Why don't you never clear this field?
> > 
> > The only case where it can be cleared is if we have called
> > rdma_addr_cancel(), and since this is the only place that does it and
> > immediately calls rdma_resolve_ip() again, there is no reason to ever
> > clear it.
> 
> IMHO, it is better to clear instead to rely on "the only place" semantic.

Then the code looks really silly:

	if (id_priv->used_resolve_ip) {
		rdma_addr_cancel(&id->route.addr.dev_addr);
                id_priv->used_resolve_ip = 0;
        }
        id_priv->used_resolve_ip = 1;

Jason

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH rc] RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests
  2021-09-23 11:45       ` Jason Gunthorpe
@ 2021-09-23 18:15         ` Leon Romanovsky
  2021-09-23 20:03           ` Jason Gunthorpe
  0 siblings, 1 reply; 10+ messages in thread
From: Leon Romanovsky @ 2021-09-23 18:15 UTC (permalink / raw)
  To: Jason Gunthorpe; +Cc: Dmitry Vyukov, linux-rdma, syzbot+dc3dfba010d7671e05f5

On Thu, Sep 23, 2021 at 08:45:57AM -0300, Jason Gunthorpe wrote:
> On Thu, Sep 23, 2021 at 08:49:06AM +0300, Leon Romanovsky wrote:
> > On Wed, Sep 22, 2021 at 11:41:19AM -0300, Jason Gunthorpe wrote:
> > > On Wed, Sep 22, 2021 at 11:01:39AM +0300, Leon Romanovsky wrote:
> > > 
> > > > > +			/* The FSM can return back to RDMA_CM_ADDR_BOUND after
> > > > > +			 * rdma_resolve_ip() is called, eg through the error
> > > > > +			 * path in addr_handler. If this happens the existing
> > > > > +			 * request must be canceled before issuing a new one.
> > > > > +			 */
> > > > > +			if (id_priv->used_resolve_ip)
> > > > > +				rdma_addr_cancel(&id->route.addr.dev_addr);
> > > > > +			else
> > > > > +				id_priv->used_resolve_ip = 1;
> > > > 
> > > > Why don't you never clear this field?
> > > 
> > > The only case where it can be cleared is if we have called
> > > rdma_addr_cancel(), and since this is the only place that does it and
> > > immediately calls rdma_resolve_ip() again, there is no reason to ever
> > > clear it.
> > 
> > IMHO, it is better to clear instead to rely on "the only place" semantic.
> 
> Then the code looks really silly:
> 
> 	if (id_priv->used_resolve_ip) {
> 		rdma_addr_cancel(&id->route.addr.dev_addr);
>                 id_priv->used_resolve_ip = 0;
>         }
>         id_priv->used_resolve_ip = 1;

So write comment why you don't need to clear used_resolve_ip, but don't
leave it as it is now, where readers need to guess.

Thanks

> 
> Jason

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH rc] RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests
  2021-09-23 18:15         ` Leon Romanovsky
@ 2021-09-23 20:03           ` Jason Gunthorpe
  2021-09-23 23:17             ` Leon Romanovsky
  0 siblings, 1 reply; 10+ messages in thread
From: Jason Gunthorpe @ 2021-09-23 20:03 UTC (permalink / raw)
  To: Leon Romanovsky; +Cc: Dmitry Vyukov, linux-rdma, syzbot+dc3dfba010d7671e05f5

On Thu, Sep 23, 2021 at 09:15:44PM +0300, Leon Romanovsky wrote:
> On Thu, Sep 23, 2021 at 08:45:57AM -0300, Jason Gunthorpe wrote:
> > On Thu, Sep 23, 2021 at 08:49:06AM +0300, Leon Romanovsky wrote:
> > > On Wed, Sep 22, 2021 at 11:41:19AM -0300, Jason Gunthorpe wrote:
> > > > On Wed, Sep 22, 2021 at 11:01:39AM +0300, Leon Romanovsky wrote:
> > > > 
> > > > > > +			/* The FSM can return back to RDMA_CM_ADDR_BOUND after
> > > > > > +			 * rdma_resolve_ip() is called, eg through the error
> > > > > > +			 * path in addr_handler. If this happens the existing
> > > > > > +			 * request must be canceled before issuing a new one.
> > > > > > +			 */
> > > > > > +			if (id_priv->used_resolve_ip)
> > > > > > +				rdma_addr_cancel(&id->route.addr.dev_addr);
> > > > > > +			else
> > > > > > +				id_priv->used_resolve_ip = 1;
> > > > > 
> > > > > Why don't you never clear this field?
> > > > 
> > > > The only case where it can be cleared is if we have called
> > > > rdma_addr_cancel(), and since this is the only place that does it and
> > > > immediately calls rdma_resolve_ip() again, there is no reason to ever
> > > > clear it.
> > > 
> > > IMHO, it is better to clear instead to rely on "the only place" semantic.
> > 
> > Then the code looks really silly:
> > 
> > 	if (id_priv->used_resolve_ip) {
> > 		rdma_addr_cancel(&id->route.addr.dev_addr);
> >                 id_priv->used_resolve_ip = 0;
> >         }
> >         id_priv->used_resolve_ip = 1;
> 
> So write comment why you don't need to clear used_resolve_ip, but don't
> leave it as it is now, where readers need to guess.
>

I think it is a bit wordy, but I put this:

			/*
			 * The FSM can return back to RDMA_CM_ADDR_BOUND after
			 * rdma_resolve_ip() is called, eg through the error
			 * path in addr_handler(). If this happens the existing
			 * request must be canceled before issuing a new one.
			 * Since canceling a request is a bit slow and this
			 * oddball path is rare, keep track once a request has
			 * been issued. The track turns out to be a permanent
			 * state since this is the only cancel as it is
			 * immediately before rdma_resolve_ip().
			 */

And into for-rc

Jason

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH rc] RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests
  2021-09-23 20:03           ` Jason Gunthorpe
@ 2021-09-23 23:17             ` Leon Romanovsky
  0 siblings, 0 replies; 10+ messages in thread
From: Leon Romanovsky @ 2021-09-23 23:17 UTC (permalink / raw)
  To: Jason Gunthorpe; +Cc: Dmitry Vyukov, linux-rdma, syzbot+dc3dfba010d7671e05f5

On Thu, Sep 23, 2021 at 05:03:58PM -0300, Jason Gunthorpe wrote:
> On Thu, Sep 23, 2021 at 09:15:44PM +0300, Leon Romanovsky wrote:
> > On Thu, Sep 23, 2021 at 08:45:57AM -0300, Jason Gunthorpe wrote:
> > > On Thu, Sep 23, 2021 at 08:49:06AM +0300, Leon Romanovsky wrote:
> > > > On Wed, Sep 22, 2021 at 11:41:19AM -0300, Jason Gunthorpe wrote:
> > > > > On Wed, Sep 22, 2021 at 11:01:39AM +0300, Leon Romanovsky wrote:
> > > > > 
> > > > > > > +			/* The FSM can return back to RDMA_CM_ADDR_BOUND after
> > > > > > > +			 * rdma_resolve_ip() is called, eg through the error
> > > > > > > +			 * path in addr_handler. If this happens the existing
> > > > > > > +			 * request must be canceled before issuing a new one.
> > > > > > > +			 */
> > > > > > > +			if (id_priv->used_resolve_ip)
> > > > > > > +				rdma_addr_cancel(&id->route.addr.dev_addr);
> > > > > > > +			else
> > > > > > > +				id_priv->used_resolve_ip = 1;
> > > > > > 
> > > > > > Why don't you never clear this field?
> > > > > 
> > > > > The only case where it can be cleared is if we have called
> > > > > rdma_addr_cancel(), and since this is the only place that does it and
> > > > > immediately calls rdma_resolve_ip() again, there is no reason to ever
> > > > > clear it.
> > > > 
> > > > IMHO, it is better to clear instead to rely on "the only place" semantic.
> > > 
> > > Then the code looks really silly:
> > > 
> > > 	if (id_priv->used_resolve_ip) {
> > > 		rdma_addr_cancel(&id->route.addr.dev_addr);
> > >                 id_priv->used_resolve_ip = 0;
> > >         }
> > >         id_priv->used_resolve_ip = 1;
> > 
> > So write comment why you don't need to clear used_resolve_ip, but don't
> > leave it as it is now, where readers need to guess.
> >
> 
> I think it is a bit wordy, but I put this:
> 
> 			/*
> 			 * The FSM can return back to RDMA_CM_ADDR_BOUND after
> 			 * rdma_resolve_ip() is called, eg through the error
> 			 * path in addr_handler(). If this happens the existing
> 			 * request must be canceled before issuing a new one.
> 			 * Since canceling a request is a bit slow and this
> 			 * oddball path is rare, keep track once a request has
> 			 * been issued. The track turns out to be a permanent
> 			 * state since this is the only cancel as it is
> 			 * immediately before rdma_resolve_ip().
> 			 */
> 
> And into for-rc

Thanks

> 
> Jason

^ permalink raw reply	[flat|nested] 10+ messages in thread

end of thread, other threads:[~2021-09-23 23:17 UTC | newest]

Thread overview: 10+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-09-16 18:34 [PATCH rc] RDMA/cma: Ensure rdma_addr_cancel() happens before issuing more requests Jason Gunthorpe
2021-09-22  8:01 ` Leon Romanovsky
2021-09-22  9:38   ` Haakon Bugge
2021-09-22 14:44     ` Jason Gunthorpe
2021-09-22 14:41   ` Jason Gunthorpe
2021-09-23  5:49     ` Leon Romanovsky
2021-09-23 11:45       ` Jason Gunthorpe
2021-09-23 18:15         ` Leon Romanovsky
2021-09-23 20:03           ` Jason Gunthorpe
2021-09-23 23:17             ` Leon Romanovsky

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).