When an invalid (non existing) handle is used in a TPM command, that uses the resource manager interface (/dev/tpmrm0) the resource manager tries to load it from its internal cache, but fails and the tpm_dev_transmit returns an -EINVAL error to the caller. The existing async handler doesn't handle these error cases currently and the condition in the poll handler never returns mask with EPOLLIN set. The result is that the poll call blocks and the application gets stuck until the user_read_timer wakes it up after 120 sec. Change the tpm_dev_async_work function to handle error conditions returned from tpm_dev_transmit they are also reflected in the poll mask and a correct error code could passed back to the caller.
Cc: Jarkko Sakkinen jarkko@kernel.org Cc: Jason Gunthorpe jgg@ziepe.ca Cc: linux-integrity@vger.kernel.org Cc: stable@vger.kernel.org Cc: linux-kernel@vger.kernel.org Fixes: 9e1b74a63f77 ("tpm: add support for nonblocking operation") Signed-off-by: Tadeusz Struk tstruk@gmail.com --- Changed in v2: - Updated commit message with better problem description - Fixed typeos. Changed in v3: - Added a comment to tpm_dev_async_work. - Updated commit message. --- drivers/char/tpm/tpm-dev-common.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)
diff --git a/drivers/char/tpm/tpm-dev-common.c b/drivers/char/tpm/tpm-dev-common.c index c08cbb306636..50df8f09ff79 100644 --- a/drivers/char/tpm/tpm-dev-common.c +++ b/drivers/char/tpm/tpm-dev-common.c @@ -69,7 +69,13 @@ static void tpm_dev_async_work(struct work_struct *work) ret = tpm_dev_transmit(priv->chip, priv->space, priv->data_buffer, sizeof(priv->data_buffer)); tpm_put_ops(priv->chip); - if (ret > 0) { + + /* + * If ret is > 0 then tpm_dev_transmit returned the size of the + * response. If ret is < 0 then tpm_dev_transmit failed and + * returned a return code. + */ + if (ret != 0) { priv->response_length = ret; mod_timer(&priv->user_read_timer, jiffies + (120 * HZ)); }