[ldv-project] WARNING in hif_usb_alloc_rx_urbs/usb_submit_urb

[Fedor Pchelkin]

Sat, 10 Oct 2020 at 04:08:19 UTC+3, Alan Stern wrote:
 > Index: usb-devel/drivers/net/wireless/ath/ath9k/hif_usb.c
 > ===================================================================
 > --- usb-devel.orig/drivers/net/wireless/ath/ath9k/hif_usb.c
 > +++ usb-devel/drivers/net/wireless/ath/ath9k/hif_usb.c
 > @@ -1307,6 +1307,20 @@ static int ath9k_hif_usb_probe(struct us
 > struct usb_device *udev = interface_to_usbdev(interface);
 > struct hif_device_usb *hif_dev;
 > int ret = 0;
 > + struct usb_host_interface *alt;
 > + struct usb_endpoint_descriptor *epd;
 > +
 > + /* Verify the expected endpoints are present */
 > + alt = interface->cur_altsetting;
 > + if (!usb_find_int_in_endpoint(alt, &epd) ||
 > + usb_endpoint_num(epd) != USB_REG_IN_PIPE ||
 > + !usb_find_int_out_endpoint(alt, &epd) ||
 > + usb_endpoint_num(epd) != USB_REG_OUT_PIPE ||
 > + !usb_find_bulk_in_endpoint(alt, &epd) ||
 > + usb_endpoint_num(epd) != USB_WLAN_RX_PIPE ||
 > + !usb_find_bulk_out_endpoint(alt, &epd) ||
 > + usb_endpoint_num(epd) != USB_WLAN_TX_PIPE)
 > + return -ENODEV;
 >
 > if (id->driver_info == STORAGE_DEVICE)
 > return send_eject_command(interface);

We've tested the suggested patch and found a null-ptr-deref. The thing
is that usb_find_{...}_endpoint() returns zero in normal case, and
non-zero value (-ENXIO) when failed (in current patch version it is
supposed to be just opposite and sometimes a NULL epd is dereferenced).
To fix it the negation signs before usb_find_{...}_endpoint() should be
removed.

And we also think usb_find_common_endpoints(...) should be used directly
as all the scanned usb_endpoint_descriptors will be passed to it and
returned in just one call.

If you wish, I may prepare the patch myself.

Fedor