Skip to contentSkip to conversation

Two fix for dwc2 gadget driver

PATCH In review [email protected] 3 participants 18 messages Nov 30, 2015 – Dec 18, 2015

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover Two fix for dwc2 gadget driver

From: "Du, Changbin" <[email protected]>

With the first patch, enable a enabled ep will return -EBUSY. The second patch forbid queuing on disabled ep to avoid panic.

Du, Changbin (2):
usb: dwc2: add ep enabled flag to avoid double enable/disable
usb: dwc2: forbid queuing request to a disabled ep

 drivers/usb/dwc2/core.h   |  1 +
 drivers/usb/dwc2/gadget.c | 26 +++++++++++++++++++++++++-
 2 files changed, 26 insertions(+), 1 deletion(-)
Signature · 1 line
2.5.0
#1 lore raw

PATCH 1/2 usb: dwc2: add ep enabled flag to avoid double enable/disable

From: "Du, Changbin" <[email protected]>

Enabling a already enabled ep is illegal, because the ep may has trbs running. Reprogram the ep may break running transfer. So udc driver must avoid this happening by return an error -EBUSY. Gadget function driver also should avoid such things, but that is out of udc driver.

Similarly, disable a disabled ep makes no sense, but no need return an error here.

---
 drivers/usb/dwc2/core.h   |  1 +
 drivers/usb/dwc2/gadget.c | 20 +++++++++++++++++++-
 2 files changed, 20 insertions(+), 1 deletion(-)
drivers/usb/dwc2/core.h +1 -0
index a66d3cb..cf7eccd 100644
@@ -162,6 +162,7 @@ struct dwc2_hsotg_ep {
 	unsigned char           mc;
 	unsigned char           interval;
 
+	unsigned int		enabled:1;
 	unsigned int            halted:1;
 	unsigned int            periodic:1;
 	unsigned int            isochronous:1;
drivers/usb/dwc2/gadget.c +19 -1
index 0abf73c..586bbcd 100644
@@ -2423,6 +2423,7 @@ void dwc2_hsotg_core_init_disconnected(struct dwc2_hsotg *hsotg,
 	/* enable, but don't activate EP0in */
 	dwc2_writel(dwc2_hsotg_ep0_mps(hsotg->eps_out[0]->ep.maxpacket) |
 	       DXEPCTL_USBACTEP, hsotg->regs + DIEPCTL0);
+	hsotg->eps_out[0]->enabled = 1;
 
 	dwc2_hsotg_enqueue_setup(hsotg);
 
@@ -2680,6 +2681,14 @@ static int dwc2_hsotg_ep_enable(struct usb_ep *ep,
 		return -EINVAL;
 	}
 
+	spin_lock_irqsave(&hsotg->lock, flags);
+	if (hs_ep->enabled) {
+		dev_warn(hsotg->dev, "%s: ep %s already enabled\n",
+			__func__, hs_ep->name);
+		ret = -EBUSY;
+		goto error;
+	}
+
 	mps = usb_endpoint_maxp(desc);
 
 	/* note, we handle this here instead of dwc2_hsotg_set_ep_maxpacket */
@@ -2690,7 +2699,6 @@ static int dwc2_hsotg_ep_enable(struct usb_ep *ep,
 	dev_dbg(hsotg->dev, "%s: read DxEPCTL=0x%08x from 0x%08x\n",
 		__func__, epctrl, epctrl_reg);
 
-	spin_lock_irqsave(&hsotg->lock, flags);
 
 	epctrl &= ~(DXEPCTL_EPTYPE_MASK | DXEPCTL_MPS_MASK);
 	epctrl |= DXEPCTL_MPS(mps);
@@ -2806,6 +2814,8 @@ static int dwc2_hsotg_ep_enable(struct usb_ep *ep,
 	/* enable the endpoint interrupt */
 	dwc2_hsotg_ctrl_epint(hsotg, index, dir_in, 1);
 
+	hs_ep->enabled = 1;
+
 error:
 	spin_unlock_irqrestore(&hsotg->lock, flags);
 	return ret;
@@ -2835,6 +2845,11 @@ static int dwc2_hsotg_ep_disable(struct usb_ep *ep)
 	epctrl_reg = dir_in ? DIEPCTL(index) : DOEPCTL(index);
 
 	spin_lock_irqsave(&hsotg->lock, flags);
+	if (!hs_ep->enabled) {
+		dev_warn(hsotg->dev, "%s: ep %s already disabled\n",
+			__func__, hs_ep->name);
+		goto out;
+	}
 
 	hsotg->fifo_map &= ~(1<<hs_ep->fifo_index);
 	hs_ep->fifo_index = 0;
@@ -2854,6 +2869,9 @@ static int dwc2_hsotg_ep_disable(struct usb_ep *ep)
 	/* terminate all requests with shutdown */
 	kill_all_requests(hsotg, hs_ep, -ESHUTDOWN);
 
+	hs_ep->enabled = 0;
+
+out:
 	spin_unlock_irqrestore(&hsotg->lock, flags);
 	return 0;
 }
Signature · 1 line
2.5.0
#2 lore raw

Re: [PATCH 1/2] usb: dwc2: add ep enabled flag to avoid double enable/disable

Hi,

Quoted 41 lines · [email protected] writes:

From: "Du, Changbin" <[email protected]>

Enabling a already enabled ep is illegal, because the ep may has trbs running. Reprogram the ep may break running transfer. So udc driver must avoid this happening by return an error -EBUSY. Gadget function driver also should avoid such things, but that is out of udc driver.

Similarly, disable a disabled ep makes no sense, but no need return an error here.

---
 drivers/usb/dwc2/core.h   |  1 +
 drivers/usb/dwc2/gadget.c | 20 +++++++++++++++++++-
 2 files changed, 20 insertions(+), 1 deletion(-)
drivers/usb/dwc2/core.h +1 -0
index a66d3cb..cf7eccd 100644
@@ -162,6 +162,7 @@ struct dwc2_hsotg_ep {
 	unsigned char           mc;
 	unsigned char           interval;
 
+	unsigned int		enabled:1;
 	unsigned int            halted:1;
 	unsigned int            periodic:1;
 	unsigned int            isochronous:1;
drivers/usb/dwc2/gadget.c +1 -0
index 0abf73c..586bbcd 100644
@@ -2423,6 +2423,7 @@ void dwc2_hsotg_core_init_disconnected(struct dwc2_hsotg *hsotg,
 	/* enable, but don't activate EP0in */
 	dwc2_writel(dwc2_hsotg_ep0_mps(hsotg->eps_out[0]->ep.maxpacket) |
 	       DXEPCTL_USBACTEP, hsotg->regs + DIEPCTL0);
+	hsotg->eps_out[0]->enabled = 1;
 
 	dwc2_hsotg_enqueue_setup(hsotg);
 

drivers/usb/dwc2/gadget.c static int dwc2_hsotg_ep_enable(struct usb_ep *ep, · line 2687

Quoted 7 lines
drivers/usb/dwc2/gadget.c +4 -0
@@ -2680,6 +2681,14 @@ static int dwc2_hsotg_ep_enable(struct usb_ep *ep,
 		return -EINVAL;
 	}
 
+	spin_lock_irqsave(&hsotg->lock, flags);
+	if (hs_ep->enabled) {
+		dev_warn(hsotg->dev, "%s: ep %s already enabled\n",
+			__func__, hs_ep->name);

this is a rather serious condition. I'd rather use dev_WARN_ONCE():

	if (dev_WARN_ONCE(hsotg->dev, hs_ep->enabled,
           	"ep %s already enabled\n", hs_ep->name)) {
Signature · 1 line
balbi
#3 lore raw

RE: [PATCH 1/2] usb: dwc2: add ep enabled flag to avoid double enable/disable

Hi, Balbi,
Please ignore this patch set. Because we have a fix in gadget API layer. [PATCH] usb: gadget: forbid queuing request to a disabled ep

Quoted 57 lines

drivers/usb/dwc2/gadget.c void dwc2_hsotg_core_init_disconnected(struct

Quoted 31 lines

Enabling an already enabled ep is illegal, because the ep may has trbs running. Reprogram the ep may break running transfer. So udc driver must avoid this happening by return an error -EBUSY. Gadget function driver also should avoid such things, but that is out of udc driver.

Similarly, disable a disabled ep makes no sense, but no need return an error here.

---
 drivers/usb/dwc2/core.h   |  1 +
 drivers/usb/dwc2/gadget.c | 20 +++++++++++++++++++-
 2 files changed, 20 insertions(+), 1 deletion(-)
drivers/usb/dwc2/core.h +1 -0
index a66d3cb..cf7eccd 100644
@@ -162,6 +162,7 @@ struct dwc2_hsotg_ep {
 	unsigned char           mc;
 	unsigned char           interval;
 
+	unsigned int		enabled:1;
 	unsigned int            halted:1;
 	unsigned int            periodic:1;
 	unsigned int            isochronous:1;
drivers/usb/dwc2/gadget.c +0 -0
index 0abf73c..586bbcd 100644
@@ -2423,6 +2423,7 @@ void dwc2_hsotg_core_init_disconnected(struct

dwc2_hsotg *hsotg,

drivers/usb/dwc2/gadget.c static int dwc2_hsotg_ep_enable(struct usb_ep

Quoted 9 lines
drivers/usb/dwc2/gadget.c +0 -0
 	/* enable, but don't activate EP0in */
 	dwc2_writel(dwc2_hsotg_ep0_mps(hsotg->eps_out[0]-

ep.maxpacket) |

drivers/usb/dwc2/gadget.c +1 -0
 	       DXEPCTL_USBACTEP, hsotg->regs + DIEPCTL0);
+	hsotg->eps_out[0]->enabled = 1;
 
 	dwc2_hsotg_enqueue_setup(hsotg);
@@ -2680,6 +2681,14 @@ static int dwc2_hsotg_ep_enable(struct usb_ep

*ep,

drivers/usb/dwc2/gadget.c static int dwc2_hsotg_ep_enable(struct usb_ep *ep, · line 2687

Quoted 7 lines
drivers/usb/dwc2/gadget.c +4 -0
 		return -EINVAL;
 	}
 
+	spin_lock_irqsave(&hsotg->lock, flags);
+	if (hs_ep->enabled) {
+		dev_warn(hsotg->dev, "%s: ep %s already enabled\n",
+			__func__, hs_ep->name);

this is a rather serious condition. I'd rather use dev_WARN_ONCE():

	if (dev_WARN_ONCE(hsotg->dev, hs_ep->enabled,
           	"ep %s already enabled\n", hs_ep->name)) {

--
balbi

#4 lore raw

PATCH 2/2 usb: dwc2: forbid queuing request to a disabled ep

In reply to [email protected]

From: "Du, Changbin" <[email protected]>

Queue a request to disabled ep doesn't make sense, and induce caller make mistakes.

Here is a example for the android mtp gadget function driver. A mem corruption can happen on below senario.
1) On disconnect, mtp driver disable its EPs,
2) During send_file_work and receive_file_work, mtp queues a request to ep. (The mtp driver need improve its synchronization logic!)
3) mtp_function_unbind is invoked and all mtp requests are freed.
4) when dwc2 process the request queued on step 2, will cause kernel NULL pointer dereference exception.

---
 drivers/usb/dwc2/gadget.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/usb/dwc2/gadget.c +6 -0
index 586bbcd..4d637ab 100644
@@ -786,6 +786,12 @@ static int dwc2_hsotg_ep_queue(struct usb_ep *ep, struct usb_request *req,
 		ep->name, req, req->length, req->buf, req->no_interrupt,
 		req->zero, req->short_not_ok);
 
+	if (!hs_ep->enabled) {
+		dev_warn(hs->dev, "%s: cannot queue to disabled ep\n",
+				__func__);
+		return -ESHUTDOWN;
+	}
+
 	/* Prevent new request submission when controller is suspended */
 	if (hs->lx_state == DWC2_L2) {
 		dev_dbg(hs->dev, "%s: don't submit request while suspended\n",
Signature · 1 line
2.5.0
#5 lore raw

Re: [PATCH 2/2] usb: dwc2: forbid queuing request to a disabled ep

Hi,

Quoted 24 lines · [email protected] writes:

From: "Du, Changbin" <[email protected]>

Queue a request to disabled ep doesn't make sense, and induce caller make mistakes.

Here is a example for the android mtp gadget function driver. A mem corruption can happen on below senario.
1) On disconnect, mtp driver disable its EPs,
2) During send_file_work and receive_file_work, mtp queues a request to ep. (The mtp driver need improve its synchronization logic!)
3) mtp_function_unbind is invoked and all mtp requests are freed.
4) when dwc2 process the request queued on step 2, will cause kernel NULL pointer dereference exception.

---
 drivers/usb/dwc2/gadget.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/usb/dwc2/gadget.c +0 -0
index 586bbcd..4d637ab 100644

drivers/usb/dwc2/gadget.c static int dwc2_hsotg_ep_queue(struct usb_ep *ep, struct usb_request *req, · line 791

Quoted 6 lines
drivers/usb/dwc2/gadget.c +3 -0
@@ -786,6 +786,12 @@ static int dwc2_hsotg_ep_queue(struct usb_ep *ep, struct usb_request *req,
 		ep->name, req, req->length, req->buf, req->no_interrupt,
 		req->zero, req->short_not_ok);
 
+	if (!hs_ep->enabled) {
+		dev_warn(hs->dev, "%s: cannot queue to disabled ep\n",
+				__func__);

similar comment to previous patch:

	if (dev_WARN_ONCE(hs->dev, !hs_ep->enabled,
        	"cannot queue to disabled ep %s\n", hs_ep->name))
Quoted 14 lines
drivers/usb/dwc2/gadget.c +3 -0
+		return -ESHUTDOWN;
+	}
+
 	/* Prevent new request submission when controller is suspended */
 	if (hs->lx_state == DWC2_L2) {
 		dev_dbg(hs->dev, "%s: don't submit request while suspended\n",
Signature · 7 lines
2.5.0

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Signature · 1 line
balbi
#6 lore raw
In reply to [email protected]
Quoted 4 lines · On 11/29/2015 9:29 PM, [email protected] wrote:

From: "Du, Changbin" <[email protected]>

With the first patch, enable a enabled ep will return -EBUSY. The second patch forbid queuing on disabled ep to avoid panic.

The usb_ep->enabled flag was added in 4.4.

It looks like these same checks are also added at the API level in the usb_ep_enable() and usb_ep_disable().

In case this is bypassed we should probably add them in the gadget anyways but using the existing flag.

Regards,
John

Quoted 9 lines

Du, Changbin (2):
usb: dwc2: add ep enabled flag to avoid double enable/disable
usb: dwc2: forbid queuing request to a disabled ep

 drivers/usb/dwc2/core.h   |  1 +
 drivers/usb/dwc2/gadget.c | 26 +++++++++++++++++++++++++-
 2 files changed, 26 insertions(+), 1 deletion(-)
#7 lore raw
Quoted 9 lines · On 11/29/2015 9:29 PM, [email protected] wrote:
Quoted 4 lines

From: "Du, Changbin" <[email protected]>

With the first patch, enable a enabled ep will return -EBUSY. The second patch forbid queuing on disabled ep to avoid panic.

The usb_ep->enabled flag was added in 4.4.

Quoted 9 lines

It looks like these same checks are also added at the API level in the usb_ep_enable() and usb_ep_disable().

In case this is bypassed we should probably add them in the gadget anyways but using the existing flag.

Regards,
John

Hmm, just learnt the flag on gadget API layer. And I just see usb_ep_enable return success if it is already enabled. But I think it should return an error to inform the caller. Because the ep configuration may probably be changed. In this case, usb_ep_enable will do different behavior.

Hmm, the usb_ep_queue doesn't check the enabled flag. Should be added. Let me have a try.

Best Regards,
Changbin

#8 lore raw

PATCH 1/1 usb: gadget: forbid queuing request to a disabled ep

From: "Du, Changbin" <[email protected]>

Queue a request to disabled ep doesn't make sense, and induce caller make mistakes.

Here is a example for the android mtp gadget function driver. A mem corruption can happen on below senario.
1) On disconnect, mtp driver disable its EPs,
2) During send_file_work and receive_file_work, mtp queues a request to ep. (The mtp driver need improve its synchronization logic!)
3) mtp_function_unbind is invoked and all mtp requests are freed.
4) when udc process the request queued on step 2, will cause kernel NULL pointer dereference exception.

---
This patch is seprated from below patches because gadget layer has added the 'enabled' flag in v4.4. so abandon it and submit new one. [PATCH 0/2] Two fix for dwc2 gadget driver
usb: dwc2: add ep enabled flag to avoid double enable/disable
usb: dwc2: forbid queuing request to a disabled ep

---
 include/linux/usb/gadget.h | 3 +++
 1 file changed, 3 insertions(+)
include/linux/usb/gadget.h +3 -0
index 3d583a1..d813bd2 100644
@@ -402,6 +402,9 @@ static inline void usb_ep_free_request(struct usb_ep *ep,
 static inline int usb_ep_queue(struct usb_ep *ep,
 			       struct usb_request *req, gfp_t gfp_flags)
 {
+	if (!ep->enabled)
+		return -ESHUTDOWN;
+
 	return ep->ops->queue(ep, req, gfp_flags);
 }
 
Signature · 1 line
2.5.0
#9 lore raw

Re: [PATCH] usb: gadget: forbid queuing request to a disabled ep

Hi,

Quoted 31 lines · [email protected] writes:

From: "Du, Changbin" <[email protected]>

Queue a request to disabled ep doesn't make sense, and induce caller make mistakes.

Here is a example for the android mtp gadget function driver. A mem corruption can happen on below senario.
1) On disconnect, mtp driver disable its EPs,
2) During send_file_work and receive_file_work, mtp queues a request to ep. (The mtp driver need improve its synchronization logic!)
3) mtp_function_unbind is invoked and all mtp requests are freed.
4) when udc process the request queued on step 2, will cause kernel NULL pointer dereference exception.

---
This patch is seprated from below patches because gadget layer has added the 'enabled' flag in v4.4. so abandon it and submit new one. [PATCH 0/2] Two fix for dwc2 gadget driver
usb: dwc2: add ep enabled flag to avoid double enable/disable
usb: dwc2: forbid queuing request to a disabled ep

---
 include/linux/usb/gadget.h | 3 +++
 1 file changed, 3 insertions(+)
include/linux/usb/gadget.h +0 -0
index 3d583a1..d813bd2 100644

include/linux/usb/gadget.h static inline void usb_ep_free_request(struct usb_ep *ep, · line 406

Quoted 5 lines
include/linux/usb/gadget.h +2 -0
@@ -402,6 +402,9 @@ static inline void usb_ep_free_request(struct usb_ep *ep,
 static inline int usb_ep_queue(struct usb_ep *ep,
 			       struct usb_request *req, gfp_t gfp_flags)
 {
+	if (!ep->enabled)
+		return -ESHUTDOWN;

same warn here:

	if (WARN_ON_ONCE(!ep->enabled))
        	return -ESHUTDOWN;
Quoted 7 lines
include/linux/usb/gadget.h +1 -0
+
 	return ep->ops->queue(ep, req, gfp_flags);
 }
 
Signature · 1 line
2.5.0
Signature · 1 line
balbi
#10 lore raw

PATCH v2 1/1 usb: gadget: forbid queuing request to a disabled ep

From: "Du, Changbin" <[email protected]>

Queue a request to disabled ep doesn't make sense, and induce caller make mistakes.

Here is a example for the android mtp gadget function driver. A mem corruption can happen on below senario.
1) On disconnect, mtp driver disable its EPs,
2) During send_file_work and receive_file_work, mtp queues a request to ep. (The mtp driver need improve its synchronization logic!)
3) mtp_function_unbind is invoked and all mtp requests are freed.
4) when udc process the request queued on step 2, will cause kernel NULL pointer dereference exception.

---
change from v1: add WARN_ON_ONCE message.

---
 include/linux/usb/gadget.h | 3 +++
 1 file changed, 3 insertions(+)
include/linux/usb/gadget.h +3 -0
index 3d583a1..b566a4b 100644
@@ -402,6 +402,9 @@ static inline void usb_ep_free_request(struct usb_ep *ep,
 static inline int usb_ep_queue(struct usb_ep *ep,
 			       struct usb_request *req, gfp_t gfp_flags)
 {
+	if (WARN_ON_ONCE(!ep->enabled))
+		return -ESHUTDOWN;
+
 	return ep->ops->queue(ep, req, gfp_flags);
 }
 
Signature · 1 line
2.5.0
#11 lore raw

RE: [PATCH v2] usb: gadget: forbid queuing request to a disabled ep

Quoted 16 lines
include/linux/usb/gadget.h +3 -1
index 3d583a1..b566a4b 100644
@@ -402,6 +402,9 @@ static inline void usb_ep_free_request(struct usb_ep
 *ep,
 static inline int usb_ep_queue(struct usb_ep *ep,
 			       struct usb_request *req, gfp_t gfp_flags)
 {
+	if (WARN_ON_ONCE(!ep->enabled))
+		return -ESHUTDOWN;
+
 	return ep->ops->queue(ep, req, gfp_flags);
 }
 
--
Quoted 1 line

2.5.0

With this patch, ep0 transfer breaks. it because the 'enabled' of ep0 is not set. Ep0 is not enabled by usb_ep_enable, but in UDC driver. So there need another patch to set ep0's flag also.

#12 lore raw

RE: [PATCH v2] usb: gadget: forbid queuing request to a disabled ep

Hi,

Quoted 15 lines · Changbin
Quoted 15 lines
include/linux/usb/gadget.h +3 -0
index 3d583a1..b566a4b 100644
@@ -402,6 +402,9 @@ static inline void usb_ep_free_request(struct usb_ep
 *ep,
 static inline int usb_ep_queue(struct usb_ep *ep,
 			       struct usb_request *req, gfp_t gfp_flags)
 {
+	if (WARN_ON_ONCE(!ep->enabled))
+		return -ESHUTDOWN;
+
 	return ep->ops->queue(ep, req, gfp_flags);
 }
Quoted 6 lines
Quoted 2 lines

--
2.5.0

With this patch, ep0 transfer breaks. it because the 'enabled' of ep0 is not set. Ep0 is not enabled by usb_ep_enable, but in UDC driver. So there need another patch to set ep0's flag also.

yeah, we don't like regressions :-) So the fix should come before $subject to avoid a regression.

Signature · 1 line
balbi
#13 lore raw

RE: [PATCH v2] usb: gadget: forbid queuing request to a disabled ep

Quoted 11 lines
Quoted 5 lines
Quoted 1 line

2.5.0

With this patch, ep0 transfer breaks. it because the 'enabled' of ep0 is not set. Ep0 is not enabled by usb_ep_enable, but in UDC driver. So there need another patch to set ep0's flag also.

yeah, we don't like regressions :-) So the fix should come before $subject to avoid a regression.

--
balbi

It is hard to determine if ep0 is enabled or not in gadget API layer. Because it is controlled by udc driver, it may enable it at pullup, vbussession... But here, we can ignore for control-ep, considering it always enabled during usb session.

#14 lore raw

PATCH v3 1/1 usb: gadget: forbid queuing request to a disabled ep

In reply to Felipe Balbi

From: "Du, Changbin" <[email protected]>

Queue a request to disabled ep doesn't make sense, and induce caller make mistakes.

Here is a example for the android mtp gadget function driver. A mem corruption can happen on below senario.
1) On disconnect, mtp driver disable its EPs,
2) During send_file_work and receive_file_work, mtp queues a request to ep. (The mtp driver need improve its synchronization logic!)
3) mtp_function_unbind is invoked and all mtp requests are freed.
4) when udc process the request queued on step 2, will cause kernel NULL pointer dereference exception.

---
change from v2: igonre ep0 as it always enabled during usb session. change from v1: add WARN_ON_ONCE message.

---
 include/linux/usb/gadget.h | 3 +++
 1 file changed, 3 insertions(+)
include/linux/usb/gadget.h +3 -0
index 3d583a1..0c5d9ea 100644
@@ -402,6 +402,9 @@ static inline void usb_ep_free_request(struct usb_ep *ep,
 static inline int usb_ep_queue(struct usb_ep *ep,
 			       struct usb_request *req, gfp_t gfp_flags)
 {
+	if (WARN_ON_ONCE(!ep->enabled && !ep->address))
+		return -ESHUTDOWN;
+
 	return ep->ops->queue(ep, req, gfp_flags);
 }
 
Signature · 1 line
2.5.0
#15 lore raw

Re: [PATCH v3] usb: gadget: forbid queuing request to a disabled ep

Hi,

Quoted 27 lines · changbin.du(a)intel.com

From: "Du, Changbin" <[email protected]>

Queue a request to disabled ep doesn't make sense, and induce caller make mistakes.

Here is a example for the android mtp gadget function driver. A mem corruption can happen on below senario.
1) On disconnect, mtp driver disable its EPs,
2) During send_file_work and receive_file_work, mtp queues a request to ep. (The mtp driver need improve its synchronization logic!)
3) mtp_function_unbind is invoked and all mtp requests are freed.
4) when udc process the request queued on step 2, will cause kernel NULL pointer dereference exception.

---
change from v2: igonre ep0 as it always enabled during usb session. change from v1: add WARN_ON_ONCE message.

---
 include/linux/usb/gadget.h | 3 +++
 1 file changed, 3 insertions(+)
include/linux/usb/gadget.h +0 -0
index 3d583a1..0c5d9ea 100644

include/linux/usb/gadget.h static inline void usb_ep_free_request(struct usb_ep *ep, · line 405

Quoted 4 lines
include/linux/usb/gadget.h +1 -0
@@ -402,6 +402,9 @@ static inline void usb_ep_free_request(struct usb_ep *ep,
 static inline int usb_ep_queue(struct usb_ep *ep,
 			       struct usb_request *req, gfp_t gfp_flags)
 {
+	if (WARN_ON_ONCE(!ep->enabled && !ep->address))

this will only trigger for a disabled ep0. Are you testing any of your patches at all ?

Signature · 1 line
balbi
#16 lore raw

RE: [PATCH v3] usb: gadget: forbid queuing request to a disabled ep

Quoted 15 lines

include/linux/usb/gadget.h static inline void usb_ep_free_request(struct

Quoted 6 lines
include/linux/usb/gadget.h +0 -0
index 3d583a1..0c5d9ea 100644
@@ -402,6 +402,9 @@ static inline void usb_ep_free_request(struct

usb_ep *ep,

include/linux/usb/gadget.h static inline void usb_ep_free_request(struct usb_ep *ep, · line 405

Quoted 4 lines
include/linux/usb/gadget.h +1 -0
 static inline int usb_ep_queue(struct usb_ep *ep,
 			       struct usb_request *req, gfp_t gfp_flags)
 {
+	if (WARN_ON_ONCE(!ep->enabled && !ep->address))

this will only trigger for a disabled ep0. Are you testing any of your patches at all ?

Quoted 2 lines

--
balbi

Oops, I sent a wrong patch. I will send right patch again as v4, very sorry for this. The right patch has been verified on 3.14 by back-porting related 1 patch.

#17 lore raw

PATCH v4 1/1 usb: gadget: forbid queuing request to a disabled ep

In reply to Felipe Balbi

From: "Du, Changbin" <[email protected]>

Queue a request to disabled ep doesn't make sense, and induce caller make mistakes.

Here is a example for the android mtp gadget function driver. A mem corruption can happen on below senario.
1) On disconnect, mtp driver disable its EPs,
2) During send_file_work and receive_file_work, mtp queues a request to ep. (The mtp driver need improve its synchronization logic!)
3) mtp_function_unbind is invoked and all mtp requests are freed.
4) when udc process the request queued on step 2, will cause kernel NULL pointer dereference exception.

---
change from v3: fix v3's error.
change from v2: igonre ep0 as it always enabled during usb session. change from v1: add WARN_ON_ONCE message.

---
 include/linux/usb/gadget.h | 3 +++
 1 file changed, 3 insertions(+)
include/linux/usb/gadget.h +3 -0
index 3d583a1..fc78687 100644
@@ -402,6 +402,9 @@ static inline void usb_ep_free_request(struct usb_ep *ep,
 static inline int usb_ep_queue(struct usb_ep *ep,
 			       struct usb_request *req, gfp_t gfp_flags)
 {
+	if (WARN_ON_ONCE(!ep->enabled && ep->address))
+		return -ESHUTDOWN;
+
 	return ep->ops->queue(ep, req, gfp_flags);
 }
 
Signature · 1 line
2.5.0