On Sun, Apr 11, 2021 at 12:04:25AM +0300, Michael Zaidman wrote:

On Sat, Apr 10, 2021 at 06:37:13PM +0300, Dan Carpenter wrote:

quoted

On Sat, Apr 10, 2021 at 03:27:29PM +0300, Michael Zaidman wrote:

quoted

On Fri, Apr 09, 2021 at 03:32:06PM +0300, Dan Carpenter wrote:

quoted

Hello Michael Zaidman,

The patch 6a82582d9fa4: "HID: ft260: add usb hid to i2c host bridge
driver" from Feb 19, 2021, leads to the following static checker
warning:

	drivers/hid/hid-ft260.c:441 ft260_smbus_write()
	error: '__memcpy()' '&rep->data[1]' too small (59 vs 255)

drivers/hid/hid-ft260.c
   423  static int ft260_smbus_write(struct ft260_device *dev, u8 addr, u8 cmd,
   424                               u8 *data, u8 data_len, u8 flag)
   425  {
   426          int ret = 0;
   427          int len = 4;
   428  
   429          struct ft260_i2c_write_request_report *rep =
   430                  (struct ft260_i2c_write_request_report *)dev->write_buf;
   431  
   432          rep->address = addr;
   433          rep->data[0] = cmd;
   434          rep->length = data_len + 1;
   435          rep->flag = flag;
   436          len += rep->length;
   437  
   438          rep->report = FT260_I2C_DATA_REPORT_ID(len);
   439  
   440          if (data_len > 0)
   441                  memcpy(&rep->data[1], data, data_len);
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Smatch says that this can be called from the i2cdev_ioctl_smbus()
function.

Hi Dan,

This is an example of a false-positive static checker warning.

The maximum data size that the i2cdev_ioctl_smbus() can pass to the
i2c_smbus_xfer() is sizeof(data->block) which is (I2C_SMBUS_BLOCK_MAX + 2)
or 34 bytes. Thus, no need to check the data_len against 59 here.

quoted

i2cdev_ioctl_smbus()
  --> i2c_smbus_xfer
      --> __i2c_smbus_xfer
          --> ft260_smbus_xfer
              --> ft260_smbus_write

It's actually me who misunderstood the Smatch warning.  Smatch is not
complaining about data_len, it's data->block[0] which is user
controlled and only for the I2C_SMBUS_I2C_BLOCK_DATA command.

The call tree is the same.  I've looked at it again.  Here is how
i2cdev_ioctl_smbus() looks like:

drivers/i2c/i2c-dev.c
   355                  return -EINVAL;
   356          }
   357  
   358          if ((size == I2C_SMBUS_BYTE_DATA) ||
   359              (size == I2C_SMBUS_BYTE))
   360                  datasize = sizeof(data->byte);
   361          else if ((size == I2C_SMBUS_WORD_DATA) ||
   362                   (size == I2C_SMBUS_PROC_CALL))
   363                  datasize = sizeof(data->word);
   364          else /* size == smbus block, i2c block, or block proc. call */
   365                  datasize = sizeof(data->block);
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

   366  
   367          if ((size == I2C_SMBUS_PROC_CALL) ||
   368              (size == I2C_SMBUS_BLOCK_PROC_CALL) ||
   369              (size == I2C_SMBUS_I2C_BLOCK_DATA) ||
                             ^^^^^^^^^^^^^^^^^^^^^^^^
   370              (read_write == I2C_SMBUS_WRITE)) {
   371                  if (copy_from_user(&temp, data, datasize))
                                            ^^^^
temp.block[0] is user controlled.

   372                          return -EFAULT;
   373          }
   374          if (size == I2C_SMBUS_I2C_BLOCK_BROKEN) {
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

   375                  /* Convert old I2C block commands to the new
   376                     convention. This preserves binary compatibility. */
   377                  size = I2C_SMBUS_I2C_BLOCK_DATA;
   378                  if (read_write == I2C_SMBUS_READ)
   379                          temp.block[0] = I2C_SMBUS_BLOCK_MAX;
                                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Except for size BROKEN

   380          }
   381          res = i2c_smbus_xfer(client->adapter, client->addr, client->flags,
   382                read_write, command, size, &temp);
                                                 ^^^^^

   383          if (!res && ((size == I2C_SMBUS_PROC_CALL) ||
   384                       (size == I2C_SMBUS_BLOCK_PROC_CALL) ||
   385                       (read_write == I2C_SMBUS_READ))) {
   386                  if (copy_to_user(data, &temp, datasize))
   387                          return -EFAULT;
   388          }

The rest of the call tree seems straight forward but it's possible I
have missed somewhere that checks data[0].  Here is how ft260_smbus_xfer()
looks like.

Oh, you are right. Despite that the SMbus block transaction limits the maximum
number of bytes to 32, nothing prevents a user from specifying via ioctl a larger
data size than the ft260 can handle in a single transfer.

I am going to fix it in the ft260_smbus_write (with your Signed-off-by), but
perhaps we should fix it in the first place, in the i2cdev_ioctl_smbus routine?
What do you think?

Could you just give me a Reported-by tag?  Thanks!

regards,
dan carpenter