Re: [PATCH] char: tpm: tpm-buf: Fix uninitialized return values in read helpers

[Jarkko Sakkinen <jarkko@kernel.org>]

On Thu, Apr 10, 2025 at 02:12:07PM +0530, Purva Yeshi wrote:
> On 10/04/25 13:21, Stefano Garzarella wrote:
> > On Thu, Apr 10, 2025 at 09:14:58AM +0300, Jarkko Sakkinen wrote:
> > > On Thu, Apr 10, 2025 at 02:25:36AM +0530, Purva Yeshi wrote:
> > > > Fix Smatch-detected error:
> > > > drivers/char/tpm/tpm-buf.c:208 tpm_buf_read_u8() error:
> > > > uninitialized symbol 'value'.
> > > > drivers/char/tpm/tpm-buf.c:225 tpm_buf_read_u16() error:
> > > > uninitialized symbol 'value'.
> > > > drivers/char/tpm/tpm-buf.c:242 tpm_buf_read_u32() error:
> > > > uninitialized symbol 'value'.
> > > > 
> > > > Call tpm_buf_read() to populate value but do not check its return
> > > > status. If the read fails, value remains uninitialized, causing
> > > > undefined behavior when returned or processed.
> > > > 
> > > > Initialize value to zero to ensure a defined return even if
> > > > tpm_buf_read() fails, avoiding undefined behavior from using
> > > > an uninitialized variable.
> > > 
> > > How does tpm_buf_read() fail?
> > 
> > If TPM_BUF_BOUNDARY_ERROR is set (or we are setting it), we are
> > effectively returning random stack bytes to the caller.
> > Could this be a problem?
> > 
> > If it is, maybe instead of this patch, we could set `*output` to zero in
> > the error path of tpm_buf_read(). Or return an error from tpm_buf_read()
> > so callers can return 0 or whatever they want.
> > 
> > Thanks,
> > Stefano
> > 
> 
> Hi Jarkko, Stefano,
> Thank you for the review.
> 
> I've revisited the issue and updated the implementation of tpm_buf_read() to
> zero out the *output buffer in the error paths, instead of initializing the
> return value in each caller.
> 
> static void tpm_buf_read(struct tpm_buf *buf, off_t *offset, size_t count,
> void *output)
> {
> 	off_t next_offset;
> 
> 	/* Return silently if overflow has already happened. */
> 	if (buf->flags & TPM_BUF_BOUNDARY_ERROR) {
> 		memset(output, 0, count);
> 		return;
> 	}
> 
> 	next_offset = *offset + count;
> 	if (next_offset > buf->length) {
> 		WARN(1, "tpm_buf: read out of boundary\n");
> 		buf->flags |= TPM_BUF_BOUNDARY_ERROR;
> 		memset(output, 0, count);
> 		return;
> 	}
> 
> 	memcpy(output, &buf->data[*offset], count);
> 	*offset = next_offset;
> }

Please don't touch this.

> 
> This approach ensures that output is always zeroed when the read fails,
> which avoids returning uninitialized stack values from the helper functions
> like tpm_buf_read_u8(), tpm_buf_read_u16(), and tpm_buf_read_u32().
> 
> Does this solution look acceptable for the next version of the patch?
> 
> Best regards,
> Purva Yeshi

BR, Jarkko