mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH v2] usb: dwc3: Fix race condition between concurrent dwc3_remove_requests() call paths
@ 2025-11-19 17:19 Manish Nagar
  2025-11-20  1:34 ` Thinh Nguyen
  0 siblings, 1 reply; 2+ messages in thread
From: Manish Nagar @ 2025-11-19 17:19 UTC (permalink / raw)
  To: Greg Kroah-Hartman, Thinh Nguyen; +Cc: linux-usb, linux-kernel

Addresses race condition caused by unsynchronized
execution of multiple call paths invoking `dwc3_remove_requests()`,
leading to premature freeing of USB requests and subsequent crashes.

Three distinct execution paths interact with `dwc3_remove_requests()`:
Path 1:
Triggered via `dwc3_gadget_reset_interrupt()` during USB reset
handling. The call stack includes:
- `dwc3_ep0_reset_state()`
- `dwc3_ep0_stall_and_restart()`
- `dwc3_ep0_out_start()`
- `dwc3_remove_requests()`
- `dwc3_gadget_del_and_unmap_request()`

Path 2:
Also initiated from `dwc3_gadget_reset_interrupt()`, but through
`dwc3_stop_active_transfers()`. The call stack includes:
- `dwc3_stop_active_transfers()`
- `dwc3_remove_requests()`
- `dwc3_gadget_del_and_unmap_request()`

Path 3:
Occurs independently during `adb root` execution, which triggers
USB function unbind and bind operations. The sequence includes:
- `gserial_disconnect()`
- `usb_ep_disable()`
- `dwc3_gadget_ep_disable()`
- `dwc3_remove_requests()` with `-ESHUTDOWN` status

Path 3 operates asynchronously and lacks synchronization with Paths
1 and 2. When Path 3 completes, it disables endpoints and frees 'out'
requests. If Paths 1 or 2 are still processing these requests,
accessing freed memory leads to a crash due to use-after-free conditions.

Add check for request completion for eps,  and skip processing giveback
for eps if already completed and add the request status for ep0 while
queue via_dwc3_gadget_ep0_queue.

Suggested-by: Thinh Nguyen <Thinh.Nguyen@synopsys.com>
Signed-off-by: Manish Nagar <manish.nagar@oss.qualcomm.com>
---
v2:
 - Add a check for request completion, in v1 I am avoiding this
    by wait for completion for ep0 then process the other eps.

 drivers/usb/dwc3/ep0.c    | 1 +
 drivers/usb/dwc3/gadget.c | 7 +++++++
 2 files changed, 8 insertions(+)

diff --git a/drivers/usb/dwc3/ep0.c b/drivers/usb/dwc3/ep0.c
index b4229aa13f37..e0bad5708664 100644
--- a/drivers/usb/dwc3/ep0.c
+++ b/drivers/usb/dwc3/ep0.c
@@ -94,6 +94,7 @@ static int __dwc3_gadget_ep0_queue(struct dwc3_ep *dep,
 	req->request.actual	= 0;
 	req->request.status	= -EINPROGRESS;
 	req->epnum		= dep->number;
+	req->status		= DWC3_REQUEST_STATUS_QUEUED;
 
 	list_add_tail(&req->list, &dep->pending_list);
 
diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
index 6f18b4840a25..5e4997f974dd 100644
--- a/drivers/usb/dwc3/gadget.c
+++ b/drivers/usb/dwc3/gadget.c
@@ -228,6 +228,13 @@ void dwc3_gadget_giveback(struct dwc3_ep *dep, struct dwc3_request *req,
 {
 	struct dwc3			*dwc = dep->dwc;
 
+	/*
+	 * The request might have been processed and completed while the
+	 * spinlock was released. Skip processing if already completed.
+	 */
+	if (req->status == DWC3_REQUEST_STATUS_COMPLETED)
+		return;
+
 	dwc3_gadget_del_and_unmap_request(dep, req, status);
 	req->status = DWC3_REQUEST_STATUS_COMPLETED;
 
-- 
2.25.1


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

end of thread, other threads:[~2025-11-20  1:35 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2025-11-19 17:19 [PATCH v2] usb: dwc3: Fix race condition between concurrent dwc3_remove_requests() call paths Manish Nagar
2025-11-20  1:34 ` Thinh Nguyen

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

all inboxes | Powered by JetHome®