[PATCH v8 09/10] tpm: Initialize TPM and get durations and timeouts

Add the retrieval of TPM 1.2 durations and timeouts. Since this requires
the startup of the TPM, do this for TPM 1.2 and TPM 2.

Signed-off-by: Stefan Berger <redacted>
CC: linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
CC: linux-doc-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
CC: linux-api-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
---
 drivers/char/tpm/tpm_vtpm_proxy.c | 95 +++++++++++++++++++++++++++++++++++----
 1 file changed, 86 insertions(+), 9 deletions(-)

diff --git a/drivers/char/tpm/tpm_vtpm_proxy.c b/drivers/char/tpm/tpm_vtpm_proxy.c
index 2bb2c8c..7fd686b 100644
--- a/drivers/char/tpm/tpm_vtpm_proxy.c
+++ b/drivers/char/tpm/tpm_vtpm_proxy.c

@@ -45,8 +45,11 @@ struct proxy_dev {
 	size_t req_len;              /* length of queued TPM request */
 	size_t resp_len;             /* length of queued TPM response */
 	u8 buffer[TPM_BUFSIZE];      /* request/response buffer */
+
+	struct work_struct work;     /* task that retrieves TPM timeouts */
 };
 
+static struct workqueue_struct *workqueue;
 
 static void vtpm_proxy_delete_device(struct proxy_dev *proxy_dev);
 

@@ -67,6 +70,15 @@ static ssize_t vtpm_proxy_fops_read(struct file *filp, char __user *buf,
 	size_t len;
 	int sig, rc;
 
+	mutex_lock(&proxy_dev->buf_lock);
+
+	if (!(proxy_dev->state & STATE_OPENED_FLAG)) {
+		mutex_unlock(&proxy_dev->buf_lock);
+		return -EPIPE;
+	}
+
+	mutex_unlock(&proxy_dev->buf_lock);
+
 	sig = wait_event_interruptible(proxy_dev->wq, proxy_dev->req_len != 0);
 	if (sig)
 		return -EINTR;

@@ -110,6 +122,11 @@ static ssize_t vtpm_proxy_fops_write(struct file *filp, const char __user *buf,
 
 	mutex_lock(&proxy_dev->buf_lock);
 
+	if (!(proxy_dev->state & STATE_OPENED_FLAG)) {
+		mutex_unlock(&proxy_dev->buf_lock);
+		return -EPIPE;
+	}
+
 	if (count > sizeof(proxy_dev->buffer) ||
 	    !(proxy_dev->state & STATE_WAIT_RESPONSE_FLAG)) {
 		mutex_unlock(&proxy_dev->buf_lock);

@@ -154,6 +171,9 @@ static unsigned int vtpm_proxy_fops_poll(struct file *filp, poll_table *wait)
 	if (proxy_dev->req_len)
 		ret |= POLLIN | POLLRDNORM;
 
+	if (!(proxy_dev->state & STATE_OPENED_FLAG))
+		ret |= POLLHUP;
+
 	mutex_unlock(&proxy_dev->buf_lock);
 
 	return ret;

@@ -341,6 +361,55 @@ static const struct tpm_class_ops vtpm_proxy_tpm_ops = {
 };
 
 /*
+ * Code related to the startup of the TPM 2 and startup of TPM 1.2 +
+ * retrieval of timeouts and durations.
+ */
+
+static void vtpm_proxy_work(struct work_struct *work)
+{
+	struct proxy_dev *proxy_dev = container_of(work, struct proxy_dev,
+						   work);
+	int rc;
+
+	if (proxy_dev->flags & VTPM_PROXY_FLAG_TPM2)
+		rc = tpm2_startup(proxy_dev->chip, TPM2_SU_CLEAR);
+	else
+		rc = tpm_get_timeouts(proxy_dev->chip);
+
+	if (rc)
+		goto err;
+
+	rc = tpm_chip_register(proxy_dev->chip);
+	if (rc)
+		goto err;
+
+	return;
+
+err:
+	vtpm_proxy_fops_undo_open(proxy_dev);
+}
+
+/*
+ * vtpm_proxy_work_stop: make sure the work has finished
+ *
+ * This function is useful when user space closed the fd
+ * while the driver still determines timeouts.
+ */
+static void vtpm_proxy_work_stop(struct proxy_dev *proxy_dev)
+{
+	vtpm_proxy_fops_undo_open(proxy_dev);
+	flush_work(&proxy_dev->work);
+}
+
+/*
+ * vtpm_proxy_work_start: Schedule the work for TPM 1.2 & 2 initialization
+ */
+static inline void vtpm_proxy_work_start(struct proxy_dev *proxy_dev)
+{
+	queue_work(workqueue, &proxy_dev->work);
+}
+
+/*
  * Code related to creation and deletion of device pairs
  */
 static struct proxy_dev *vtpm_proxy_create_proxy_dev(void)

@@ -355,6 +424,7 @@ static struct proxy_dev *vtpm_proxy_create_proxy_dev(void)
 
 	init_waitqueue_head(&proxy_dev->wq);
 	mutex_init(&proxy_dev->buf_lock);
+	INIT_WORK(&proxy_dev->work, vtpm_proxy_work);
 
 	chip = tpm_chip_alloc(NULL, &vtpm_proxy_tpm_ops);
 	if (IS_ERR(chip)) {

@@ -425,9 +495,7 @@ static struct file *vtpm_proxy_create_device(
 	if (proxy_dev->flags & VTPM_PROXY_FLAG_TPM2)
 		proxy_dev->chip->flags |= TPM_CHIP_FLAG_TPM2;
 
-	rc = tpm_chip_register(proxy_dev->chip);
-	if (rc)
-		goto err_vtpm_fput;
+	vtpm_proxy_work_start(proxy_dev);
 
 	vtpm_new_dev->fd = fd;
 	vtpm_new_dev->major = MAJOR(proxy_dev->chip->dev.devt);

@@ -436,12 +504,6 @@ static struct file *vtpm_proxy_create_device(
 
 	return file;
 
-err_vtpm_fput:
-	put_unused_fd(fd);
-	fput(file);
-
-	return ERR_PTR(rc);
-
 err_put_unused_fd:
 	put_unused_fd(fd);
 

@@ -456,6 +518,8 @@ err_delete_proxy_dev:
  */
 static void vtpm_proxy_delete_device(struct proxy_dev *proxy_dev)
 {
+	vtpm_proxy_work_stop(proxy_dev);
+
 	/*
 	 * A client may hold the 'ops' lock, so let it know that the server
 	 * side shuts down before we try to grab the 'ops' lock when

@@ -555,11 +619,24 @@ static int __init vtpm_module_init(void)
 		return rc;
 	}
 
+	workqueue = create_workqueue("tpm-vtpm");
+	if (!workqueue) {
+		pr_err("couldn't create workqueue\n");
+		rc = -ENOMEM;
+		goto err_vtpmx_cleanup;
+	}
+
 	return 0;
+
+err_vtpmx_cleanup:
+	vtpmx_cleanup();
+
+	return rc;
 }
 
 static void __exit vtpm_module_exit(void)
 {
+	destroy_workqueue(workqueue);
 	vtpmx_cleanup();
 }
 

-- 
2.4.3


------------------------------------------------------------------------------
Transform Data into Opportunity.
Accelerate data analysis in your applications with
Intel Data Analytics Acceleration Library.
Click to learn more.
http://pubads.g.doubleclick.net/gampad/clk?id=278785111&iu=/4140

On Sun, Mar 13, 2016 at 06:54:39PM -0400, Stefan Berger wrote:

What if STATE_OPENED_FLAG is set after mutex_unlock()?

Is there some scenario where STATE_OPENED_FLAG would evaluate false
at this point?

Actually I couldn't find a scenario where this check would be needed
because:

* In vtpm_proxy_work() vtpm_proxy_fops_undo_open() is called after
  sending TPM commands.
* vtpm_proxy_delete_device() calls vtpm_proxy_work_stop() as its
  first statement.

Am I ignoring something?

/Jarkko

On 03/22/2016 02:34 AM, Jarkko Sakkinen wrote:

On Sun, Mar 13, 2016 at 06:54:39PM -0400, Stefan Berger wrote:

quoted

Add the retrieval of TPM 1.2 durations and timeouts. Since this requires
the startup of the TPM, do this for TPM 1.2 and TPM 2.

Signed-off-by: Stefan Berger <redacted>
CC: linux-kernel@vger.kernel.org
CC: linux-doc@vger.kernel.org
CC: linux-api@vger.kernel.org

---
drivers/char/tpm/tpm_vtpm_proxy.c | 95 +++++++++++++++++++++++++++++++++++----
  1 file changed, 86 insertions(+), 9 deletions(-)

diff --git a/drivers/char/tpm/tpm_vtpm_proxy.c b/drivers/char/tpm/tpm_vtpm_proxy.c
index 2bb2c8c..7fd686b 100644
--- a/drivers/char/tpm/tpm_vtpm_proxy.c
+++ b/drivers/char/tpm/tpm_vtpm_proxy.c

@@ -45,8 +45,11 @@ struct proxy_dev {
  	size_t req_len;              /* length of queued TPM request */
  	size_t resp_len;             /* length of queued TPM response */
  	u8 buffer[TPM_BUFSIZE];      /* request/response buffer */
+
+	struct work_struct work;     /* task that retrieves TPM timeouts */
  };
  
+static struct workqueue_struct *workqueue;
  
  static void vtpm_proxy_delete_device(struct proxy_dev *proxy_dev);
  

@@ -67,6 +70,15 @@ static ssize_t vtpm_proxy_fops_read(struct file *filp, char __user *buf,
  	size_t len;
  	int sig, rc;
  
+	mutex_lock(&proxy_dev->buf_lock);
+
+	if (!(proxy_dev->state & STATE_OPENED_FLAG)) {
+		mutex_unlock(&proxy_dev->buf_lock);
+		return -EPIPE;
+	}
+
+	mutex_unlock(&proxy_dev->buf_lock);
+
  	sig = wait_event_interruptible(proxy_dev->wq, proxy_dev->req_len != 0);
  	if (sig)
  		return -EINTR;

What if STATE_OPENED_FLAG is set after mutex_unlock()?

This flag is only set when the file descriptor for the server side is 
created (vtpm_proxy_fops_open()). After that it can only be cleared 
(vtpm_fops_undo_open()) due to an error condition, which then indicates 
to the server side that the file descriptor is now unusable. One error 
condition can for example be the failure by the TPM emulator to respond 
to the TPM_Startup with a success in the response.

Is there some scenario where STATE_OPENED_FLAG would evaluate false
at this point?

Yes. The flag is reset in vtpm_fops_undo_open(), which is for example 
called in error conditions detected by the worker thread 
(vtpm_proxy_work()) where the server side for example didn't deliver the 
timeouts and durations or the TPM_Startup() wasn't successful.

Actually I couldn't find a scenario where this check would be needed
because:

* In vtpm_proxy_work() vtpm_proxy_fops_undo_open() is called after
   sending TPM commands.
* vtpm_proxy_delete_device() calls vtpm_proxy_work_stop() as its
   first statement.

Am I ignoring something?

Does the above explain it? The file descriptor is not closed by any 
failure condition, so it stays around but it becomes 'useless' if the 
work thread detected an error by the TPM emulator.

     Stefan

On Tue, Mar 22, 2016 at 06:54:30AM -0400, Stefan Berger wrote:

On 03/22/2016 02:34 AM, Jarkko Sakkinen wrote:

quoted

On Sun, Mar 13, 2016 at 06:54:39PM -0400, Stefan Berger wrote:

quoted

Add the retrieval of TPM 1.2 durations and timeouts. Since this requires
the startup of the TPM, do this for TPM 1.2 and TPM 2.

Signed-off-by: Stefan Berger <redacted>
CC: linux-kernel@vger.kernel.org
CC: linux-doc@vger.kernel.org
CC: linux-api@vger.kernel.org

---
drivers/char/tpm/tpm_vtpm_proxy.c | 95 +++++++++++++++++++++++++++++++++++----
 1 file changed, 86 insertions(+), 9 deletions(-)

diff --git a/drivers/char/tpm/tpm_vtpm_proxy.c b/drivers/char/tpm/tpm_vtpm_proxy.c
index 2bb2c8c..7fd686b 100644
--- a/drivers/char/tpm/tpm_vtpm_proxy.c
+++ b/drivers/char/tpm/tpm_vtpm_proxy.c

@@ -45,8 +45,11 @@ struct proxy_dev {
 	size_t req_len;              /* length of queued TPM request */
 	size_t resp_len;             /* length of queued TPM response */
 	u8 buffer[TPM_BUFSIZE];      /* request/response buffer */
+
+	struct work_struct work;     /* task that retrieves TPM timeouts */
 };
+static struct workqueue_struct *workqueue;
 static void vtpm_proxy_delete_device(struct proxy_dev *proxy_dev);

@@ -67,6 +70,15 @@ static ssize_t vtpm_proxy_fops_read(struct file *filp, char __user *buf,
 	size_t len;
 	int sig, rc;
+	mutex_lock(&proxy_dev->buf_lock);
+
+	if (!(proxy_dev->state & STATE_OPENED_FLAG)) {
+		mutex_unlock(&proxy_dev->buf_lock);
+		return -EPIPE;
+	}
+
+	mutex_unlock(&proxy_dev->buf_lock);
+
 	sig = wait_event_interruptible(proxy_dev->wq, proxy_dev->req_len != 0);
 	if (sig)
 		return -EINTR;

What if STATE_OPENED_FLAG is set after mutex_unlock()?

This flag is only set when the file descriptor for the server side is
created (vtpm_proxy_fops_open()). After that it can only be cleared
(vtpm_fops_undo_open()) due to an error condition, which then indicates to
the server side that the file descriptor is now unusable. One error
condition can for example be the failure by the TPM emulator to respond to
the TPM_Startup with a success in the response.

You take the lock two times and OPENED flag could change in-between.

Why couldn't you put the call after wait_event_* after taking the lock?

/Jarkko

On 03/29/2016 11:31 AM, Jarkko Sakkinen wrote:

On Tue, Mar 22, 2016 at 06:54:30AM -0400, Stefan Berger wrote:

quoted

On 03/22/2016 02:34 AM, Jarkko Sakkinen wrote:

quoted

On Sun, Mar 13, 2016 at 06:54:39PM -0400, Stefan Berger wrote:

quoted

Add the retrieval of TPM 1.2 durations and timeouts. Since this requires
the startup of the TPM, do this for TPM 1.2 and TPM 2.

Signed-off-by: Stefan Berger <redacted>
CC: linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
CC: linux-doc-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
CC: linux-api-u79uwXL29TY76Z2rM5mHXA@public.gmane.org

---
drivers/char/tpm/tpm_vtpm_proxy.c | 95 +++++++++++++++++++++++++++++++++++----
  1 file changed, 86 insertions(+), 9 deletions(-)

diff --git a/drivers/char/tpm/tpm_vtpm_proxy.c b/drivers/char/tpm/tpm_vtpm_proxy.c
index 2bb2c8c..7fd686b 100644
--- a/drivers/char/tpm/tpm_vtpm_proxy.c
+++ b/drivers/char/tpm/tpm_vtpm_proxy.c

@@ -45,8 +45,11 @@ struct proxy_dev {
  	size_t req_len;              /* length of queued TPM request */
  	size_t resp_len;             /* length of queued TPM response */
  	u8 buffer[TPM_BUFSIZE];      /* request/response buffer */
+
+	struct work_struct work;     /* task that retrieves TPM timeouts */
  };
+static struct workqueue_struct *workqueue;
  static void vtpm_proxy_delete_device(struct proxy_dev *proxy_dev);

@@ -67,6 +70,15 @@ static ssize_t vtpm_proxy_fops_read(struct file *filp, char __user *buf,
  	size_t len;
  	int sig, rc;
+	mutex_lock(&proxy_dev->buf_lock);
+
+	if (!(proxy_dev->state & STATE_OPENED_FLAG)) {
+		mutex_unlock(&proxy_dev->buf_lock);
+		return -EPIPE;
+	}
+
+	mutex_unlock(&proxy_dev->buf_lock);
+
  	sig = wait_event_interruptible(proxy_dev->wq, proxy_dev->req_len != 0);
  	if (sig)
  		return -EINTR;

What if STATE_OPENED_FLAG is set after mutex_unlock()?

This flag is only set when the file descriptor for the server side is
created (vtpm_proxy_fops_open()). After that it can only be cleared
(vtpm_fops_undo_open()) due to an error condition, which then indicates to
the server side that the file descriptor is now unusable. One error
condition can for example be the failure by the TPM emulator to respond to
the TPM_Startup with a success in the response.

You take the lock two times and OPENED flag could change in-between.

Why couldn't you put the call after wait_event_* after taking the lock?

Good point. Actually the test for the flag has to go into the 
wait_event_interruptible test (without mutex?!) and we have to look at 
that flag after the wait_event_interruptible again to check whether no 
more requests are coming due to the OPEN flag not being set anymore as 
result of an error condition.

I can post a v9 with that change. However, the dropping of the priv 
field from the vendor struct  in the other patch series, will again 
require changes either here or in that other series.

     Stefan