linux-um archives
 help / color / mirror / Atom feed
* [PATCH] um: vector: Don't clean up NAPI that was never set up
@ 2026-11-19 22:39 Suyuan Wang via B4 Relay
  2026-11-20  7:27 ` Anton Ivanov
  0 siblings, 1 reply; 2+ messages in thread
From: Suyuan Wang via B4 Relay @ 2026-11-19 22:39 UTC (permalink / raw)
  To: anton.ivanov, johannes, richard; +Cc: linux-kernel, linux-um

From: Suyuan Wang <stevewong@hust.edu.cn>

vector_net_open() calls vector_net_close() on any failure once
vp->fds is set, and vector_net_close() then runs napi_disable() and
netif_napi_del() unconditionally. But the NAPI context is only added
and enabled after build_transport_data() and the header buffer
allocations, so a failure in any of those disables a NAPI context that
was never added. napi_disable() locks n->dev, which is still NULL:

  ip link set vec0 up   (vec0:transport=gre,...,rx_key=1, no tx_key)
  Kernel panic - not syncing: Kernel mode fault at addr 0x910
   napi_disable+0x4e/0x9e
   vector_net_close+0x2b9/0x87d
   vector_net_open+0x105d/0x106f
   __dev_open+0x244/0x334

Record in vp->napi_enabled whether vector_net_open() got as far as
enabling NAPI, and only disable and delete it in vector_net_close()
when it did. NAPI is still enabled only after the queues and buffers
that vector_poll() uses are set up.

Found with a static analysis tool and checked by hand.

Fixes: b35507a4cfb2 ("um: Migrate vector drivers to NAPI")
Cc: stable@vger.kernel.org
Co-developed-by: Kaixuan Li <kaixuan.li@ntu.edu.sg>
Signed-off-by: Kaixuan Li <kaixuan.li@ntu.edu.sg>
Signed-off-by: Suyuan Wang <stevewong@hust.edu.cn>
---
Tested with UML (seccomp=on) on top of the um fixes branch: with a GRE
vector device missing tx_key, "ip link set vec0 up" panics without the
patch and fails cleanly with -EINVAL with it, also when repeated. A raw
vector device on the host's lo still goes up and down twice as before.
---
 arch/um/drivers/vector_kern.c | 9 +++++++--
 arch/um/drivers/vector_kern.h | 1 +
 2 files changed, 8 insertions(+), 2 deletions(-)

diff --git a/arch/um/drivers/vector_kern.c b/arch/um/drivers/vector_kern.c
index 8008278ccee7..021f14780b45 100644
--- a/arch/um/drivers/vector_kern.c
+++ b/arch/um/drivers/vector_kern.c
@@ -1137,8 +1137,12 @@ static int vector_net_close(struct net_device *dev)
 		um_free_irq(vp->tx_irq, dev);
 		vp->tx_irq = 0;
 	}
-	napi_disable(&vp->napi);
-	netif_napi_del(&vp->napi);
+	/* vector_net_open() may fail before setting up NAPI */
+	if (vp->napi_enabled) {
+		napi_disable(&vp->napi);
+		netif_napi_del(&vp->napi);
+		vp->napi_enabled = false;
+	}
 	if (vp->fds->rx_fd > 0) {
 		if (vp->bpf)
 			uml_vector_detach_bpf(vp->fds->rx_fd, vp->bpf);
@@ -1260,6 +1264,7 @@ static int vector_net_open(struct net_device *dev)
 		netif_napi_add_weight(vp->dev, &vp->napi, vector_poll, NAPI_POLL_WEIGHT);
 
 	napi_enable(&vp->napi);
+	vp->napi_enabled = true;
 
 	/* READ IRQ */
 	err = um_request_irq(
diff --git a/arch/um/drivers/vector_kern.h b/arch/um/drivers/vector_kern.h
index 5f0ceb80779d..e1b697266fa7 100644
--- a/arch/um/drivers/vector_kern.h
+++ b/arch/um/drivers/vector_kern.h
@@ -118,6 +118,7 @@ struct vector_private {
 
 	bool rexmit_scheduled;
 	bool opened;
+	bool napi_enabled;
 	bool in_write_poll;
 	bool in_error;
 






^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH] um: vector: Don't clean up NAPI that was never set up
  2026-11-19 22:39 [PATCH] um: vector: Don't clean up NAPI that was never set up Suyuan Wang via B4 Relay
@ 2026-11-20  7:27 ` Anton Ivanov
  0 siblings, 0 replies; 2+ messages in thread
From: Anton Ivanov @ 2026-11-20  7:27 UTC (permalink / raw)
  To: shyckless, johannes, richard; +Cc: linux-kernel, linux-um

On 19/11/2026 22:39, Suyuan Wang via B4 Relay wrote:
> From: Suyuan Wang <stevewong@hust.edu.cn>
> 
> vector_net_open() calls vector_net_close() on any failure once
> vp->fds is set, and vector_net_close() then runs napi_disable() and
> netif_napi_del() unconditionally. But the NAPI context is only added
> and enabled after build_transport_data() and the header buffer
> allocations, so a failure in any of those disables a NAPI context that
> was never added. napi_disable() locks n->dev, which is still NULL:
> 
>    ip link set vec0 up   (vec0:transport=gre,...,rx_key=1, no tx_key)
>    Kernel panic - not syncing: Kernel mode fault at addr 0x910
>     napi_disable+0x4e/0x9e
>     vector_net_close+0x2b9/0x87d
>     vector_net_open+0x105d/0x106f
>     __dev_open+0x244/0x334
> 
> Record in vp->napi_enabled whether vector_net_open() got as far as
> enabling NAPI, and only disable and delete it in vector_net_close()
> when it did. NAPI is still enabled only after the queues and buffers
> that vector_poll() uses are set up.
> 
> Found with a static analysis tool and checked by hand.
> 
> Fixes: b35507a4cfb2 ("um: Migrate vector drivers to NAPI")
> Cc: stable@vger.kernel.org
> Co-developed-by: Kaixuan Li <kaixuan.li@ntu.edu.sg>
> Signed-off-by: Kaixuan Li <kaixuan.li@ntu.edu.sg>
> Signed-off-by: Suyuan Wang <stevewong@hust.edu.cn>
> ---
> Tested with UML (seccomp=on) on top of the um fixes branch: with a GRE
> vector device missing tx_key, "ip link set vec0 up" panics without the
> patch and fails cleanly with -EINVAL with it, also when repeated. A raw
> vector device on the host's lo still goes up and down twice as before.
> ---
>   arch/um/drivers/vector_kern.c | 9 +++++++--
>   arch/um/drivers/vector_kern.h | 1 +
>   2 files changed, 8 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/um/drivers/vector_kern.c b/arch/um/drivers/vector_kern.c
> index 8008278ccee7..021f14780b45 100644
> --- a/arch/um/drivers/vector_kern.c
> +++ b/arch/um/drivers/vector_kern.c
> @@ -1137,8 +1137,12 @@ static int vector_net_close(struct net_device *dev)
>   		um_free_irq(vp->tx_irq, dev);
>   		vp->tx_irq = 0;
>   	}
> -	napi_disable(&vp->napi);
> -	netif_napi_del(&vp->napi);
> +	/* vector_net_open() may fail before setting up NAPI */
> +	if (vp->napi_enabled) {
> +		napi_disable(&vp->napi);
> +		netif_napi_del(&vp->napi);
> +		vp->napi_enabled = false;
> +	}
>   	if (vp->fds->rx_fd > 0) {
>   		if (vp->bpf)
>   			uml_vector_detach_bpf(vp->fds->rx_fd, vp->bpf);
> @@ -1260,6 +1264,7 @@ static int vector_net_open(struct net_device *dev)
>   		netif_napi_add_weight(vp->dev, &vp->napi, vector_poll, NAPI_POLL_WEIGHT);
>   
>   	napi_enable(&vp->napi);
> +	vp->napi_enabled = true;
>   
>   	/* READ IRQ */
>   	err = um_request_irq(
> diff --git a/arch/um/drivers/vector_kern.h b/arch/um/drivers/vector_kern.h
> index 5f0ceb80779d..e1b697266fa7 100644
> --- a/arch/um/drivers/vector_kern.h
> +++ b/arch/um/drivers/vector_kern.h
> @@ -118,6 +118,7 @@ struct vector_private {
>   
>   	bool rexmit_scheduled;
>   	bool opened;
> +	bool napi_enabled;
>   	bool in_write_poll;
>   	bool in_error;
>   
> 
> 
> 
> 
> 

Acked-By: Anton Ivanov <anton.ivanov@cambridgegreys.com>

-- 
Anton R. Ivanov
Cambridgegreys Limited. Registered in England. Company Number 10273661


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-20  7:45 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-19 22:39 [PATCH] um: vector: Don't clean up NAPI that was never set up Suyuan Wang via B4 Relay
2026-11-20  7:27 ` Anton Ivanov

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox