Re: [PATCH RFC v2 5/6] staging: r8188eu: add error handling of rtw_read32

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 8/24/21 2:33 AM, Phillip Potter wrote:
On Sun, 22 Aug 2021 at 15:36, Pavel Skripkin <paskripkin@xxxxxxxxx> wrote:
-static u32 usb_read32(struct intf_hdl *pintfhdl, u32 addr)
+static int usb_read32(struct intf_hdl *pintfhdl, u32 addr, u32 *data)
 {
        u8 requesttype;
        u16 wvalue;
        u16 len;
-       __le32 data;
+       int res;
+       __le32 tmp;
+
+       if (WARN_ON(unlikely(!data)))
+               return -EINVAL;

        requesttype = 0x01;/* read_in */

        wvalue = (u16)(addr & 0x0000ffff);
        len = 4;

-       usbctrl_vendorreq(pintfhdl, wvalue, &data, len, requesttype);
+       res = usbctrl_vendorreq(pintfhdl, wvalue, &data, len, requesttype);
+       if (res < 0) {
+               dev_err(dvobj_to_dev(pintfhdl->pintf_dev), "Failed to read 32 bytes: %d\n", res);
+       } else {
+               /* Noone cares about positive return value */
+               *data = le32_to_cpu(tmp);
+               res = 0;
+       }

-       return le32_to_cpu(data);
+       return res;
 }

Dear Pavel,

OK, found the issue with decoded stack trace after reviewing this
usb_read32 function. Your line:
res = usbctrl_vendorreq(pintfhdl, wvalue, &data, len, requesttype);

should read:
res = usbctrl_vendorreq(pintfhdl, wvalue, &tmp, len, requesttype);

With this change, the driver runs fine with no crashes/oopses. I will
explain the issue but you can probably see already, so I hope I'm not
coming across as patronising, just trying to be helpful :-)

Essentially, you are taking the address of the data function parameter
on this line with &data, a pointer to u32, which is giving you a
pointer to a pointer to u32 (u32 **) for this function parameter
variable. When passed to usbctrl_vendorreq, it is being passed to
memcpy inside this function as a void *, meaning that memcpy
subsequently overwrites the value of the memory address inside data to
point to a different location, which is problem when it is later
deferenced at:
*data = le32_to_cpu(tmp);
causing the OOPS


The most strange thing is why gcc didn't complain about different pointer types... I think, that gcc must complain about this type of not explicit casts, because 99% it's a bug.


Again big thanks for analysis :)



With regards,
Pavel Skripkin




[Index of Archives]     [Linux Driver Development]     [Linux Driver Backports]     [DMA Engine]     [Linux GPIO]     [Linux SPI]     [Video for Linux]     [Linux USB Devel]     [Linux Coverity]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [Yosemite Backpacking]
  Powered by Linux