staging: pi433: enforce tx_cfg to be set before any message can be sent

this driver relies on exposing a char device to userspace to tx messages. Every message can be sent using different trasmitter settings such so the tx_cfg must be written before sending any messages. Failing to do so will cause the message to fail silently depending on printk/dynamic_debug settings which makes it hard to troubleshoot.

This patch add a control variable that will get initialized once tx_cfg is set for the fd used when interacting with the char device.

Signed-off-by Paulo Miguel Almeida <[email protected]>

---
Patch dependency:
- This patch needs the patch below to be applied first as both tweak the same file.
https://lore.kernel.org/lkml/[email protected]/

meta-comments:
- I'm not entirely sure if -EPERM is the best error code to return here, I'm taking suggestions

---
 drivers/staging/pi433/pi433_if.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

drivers/staging/pi433/pi433_if.c +15 -1

index 051c9052aeeb..3e9913f4bc7d 100644
@@ -108,6 +108,9 @@ struct pi433_device {
 struct pi433_instance {
 	struct pi433_device	*device;
 	struct pi433_tx_cfg	tx_cfg;
+
+	/* control flags */
+	bool			tx_cfg_initialized;
 };
 
 /*-------------------------------------------------------------------------*/

@@ -823,6 +826,16 @@ pi433_write(struct file *filp, const char __user *buf,
 	if (count > MAX_MSG_SIZE)
 		return -EMSGSIZE;
 
+	/*
+	 * check if tx_cfg has been initialized otherwise we won't be able to
+	 * config the RF trasmitter correctly due to invalid settings
+	 */
+	if (!instance->tx_cfg_initialized) {
+		dev_dbg(device->dev,
+			"write: failed due to uninitialized tx_cfg");
+		return -EPERM;
+	}
+
 	/*
 	 * write the following sequence into fifo:
 	 * - tx_cfg

@@ -897,6 +910,7 @@ pi433_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 			return -EFAULT;
 		mutex_lock(&device->tx_fifo_lock);
 		memcpy(&instance->tx_cfg, &tx_cfg, sizeof(struct pi433_tx_cfg));
+		instance->tx_cfg_initialized = true;
 		mutex_unlock(&device->tx_fifo_lock);
 		break;
 	case PI433_IOC_RD_RX_CFG:

@@ -950,7 +964,7 @@ static int pi433_open(struct inode *inode, struct file *filp)
 	/* setup instance data*/
 	instance->device = device;
 	instance->tx_cfg.bit_rate = 4711;
-	// TODO: fill instance->tx_cfg;
+	instance->tx_cfg_initialized = false;
 
 	/* instance data as context */
 	filp->private_data = instance;

Signature · 1 line

2.25.4

On Sun, Jan 09, 2022 at 11:26:52AM +1300, Paulo Miguel Almeida wrote:

Hi, Paulo.
Thanks for the patch.
I have some opinion below.

Quoted 8 lines

this driver relies on exposing a char device to userspace to tx messages. Every message can be sent using different trasmitter settings such so the tx_cfg must be written before sending any messages. Failing to do so will cause the message to fail silently depending on printk/dynamic_debug settings which makes it hard to troubleshoot.

This patch add a control variable that will get initialized once tx_cfg is set for the fd used when interacting with the char device.

I don't know that adding flag is good idea. It seems that initializing to default is better than this.

Quoted 5 lines · Sidong Yang

On Sun, Jan 09, 2022 at 11:26:52AM +1300, Paulo Miguel Almeida wrote:

Hi, Paulo.
Thanks for the patch.
I have some opinion below.

thanks for taking the time to review my patch :)

Quoted 11 lines

Quoted 8 lines

this driver relies on exposing a char device to userspace to tx messages. Every message can be sent using different trasmitter settings such so the tx_cfg must be written before sending any messages. Failing to do so will cause the message to fail silently depending on printk/dynamic_debug settings which makes it hard to troubleshoot.

This patch add a control variable that will get initialized once tx_cfg is set for the fd used when interacting with the char device.

I don't know that adding flag is good idea. It seems that initializing to default is better than this.

the reasons why I thought that the flag was a good approach is because of these points:

1 - it returns an error to userspace to help the developers troubleshooting what is missing/required to make a message to be successfully tx'ed. Unfortunately, once the message is in the chip queue, there isn't much that the user can know so, from my humble point of view, any way we can avoid hard-to-debug problems, we should.

2 - rf69 work with multiple frequencies (315,433,868 and 915MHz) in which acceptance varies from region to region. Essentially, picking 1 default frequency may be ideal for one place but not right for another.

Let me know if you agree with my train of thought above and if not, share with me your point of view.

thanks,

Paulo A.

meta-comments:
- I'm not entirely sure if -EPERM is the best error code to return here, I'm taking suggestions

---
 drivers/staging/pi433/pi433_if.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

drivers/staging/pi433/pi433_if.c +3 -0

index 051c9052aeeb..3e9913f4bc7d 100644
@@ -108,6 +108,9 @@ struct pi433_device {
 struct pi433_instance {
 	struct pi433_device	*device;
 	struct pi433_tx_cfg	tx_cfg;
+
+	/* control flags */
+	bool			tx_cfg_initialized;
 };
 
 /*-------------------------------------------------------------------------*/

drivers/staging/pi433/pi433_if.c +4 -0

+		return -EPERM;
+	}
+
 	/*
 	 * write the following sequence into fifo:
 	 * - tx_cfg

@@ -897,6 +910,7 @@ pi433_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 			return -EFAULT;
 		mutex_lock(&device->tx_fifo_lock);
 		memcpy(&instance->tx_cfg, &tx_cfg, sizeof(struct pi433_tx_cfg));
+		instance->tx_cfg_initialized = true;
 		mutex_unlock(&device->tx_fifo_lock);
 		break;
 	case PI433_IOC_RD_RX_CFG:

Seems reasonable.

Quoted 3 lines · Paulo Miguel Almeida

meta-comments:
- I'm not entirely sure if -EPERM is the best error code to return here, I'm taking suggestions

Better to just return -EINVAL.

Quoted 19 lines

---
 drivers/staging/pi433/pi433_if.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

drivers/staging/pi433/pi433_if.c +3 -0

index 051c9052aeeb..3e9913f4bc7d 100644
@@ -108,6 +108,9 @@ struct pi433_device {
 struct pi433_instance {
 	struct pi433_device	*device;
 	struct pi433_tx_cfg	tx_cfg;
+
+	/* control flags */
+	bool			tx_cfg_initialized;
 };
 
 /*-------------------------------------------------------------------------*/

drivers/staging/pi433/pi433_if.c pi433_write(struct file *filp, const char __user *buf, · line 835

Quoted 10 lines

drivers/staging/pi433/pi433_if.c +7 -0

@@ -823,6 +826,16 @@ pi433_write(struct file *filp, const char __user *buf,
 	if (count > MAX_MSG_SIZE)
 		return -EMSGSIZE;
 
+	/*
+	 * check if tx_cfg has been initialized otherwise we won't be able to
+	 * config the RF trasmitter correctly due to invalid settings
+	 */
+	if (!instance->tx_cfg_initialized) {
+		dev_dbg(device->dev,
+			"write: failed due to uninitialized tx_cfg");

Use dev_notice_once() or similar. Maybe "unconfigured" instead of uninitialized?

	dev_notice_once(device->dev,
			"write: failed due to unconfigured tx_cfg (see PI433_IOC_WR_TX_CFG)");

Quoted 15 lines

drivers/staging/pi433/pi433_if.c +4 -0

+		return -EPERM;
+	}
+
 	/*
 	 * write the following sequence into fifo:
 	 * - tx_cfg

@@ -897,6 +910,7 @@ pi433_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 			return -EFAULT;
 		mutex_lock(&device->tx_fifo_lock);
 		memcpy(&instance->tx_cfg, &tx_cfg, sizeof(struct pi433_tx_cfg));
+		instance->tx_cfg_initialized = true;
 		mutex_unlock(&device->tx_fifo_lock);
 		break;
 	case PI433_IOC_RD_RX_CFG:

drivers/staging/pi433/pi433_if.c static int pi433_open(struct inode *inode, struct file *filp) · line 966

Quoted 3 lines

drivers/staging/pi433/pi433_if.c +0 -0

@@ -950,7 +964,7 @@ static int pi433_open(struct inode *inode, struct file *filp)
 	/* setup instance data*/
 	instance->device = device;
 	instance->tx_cfg.bit_rate = 4711;

This is sort of pointless because it can't work until that gets over written.

drivers/staging/pi433/pi433_if.c static int pi433_open(struct inode *inode, struct file *filp) · line 967

Quoted 2 lines

drivers/staging/pi433/pi433_if.c +1 -1

-	// TODO: fill instance->tx_cfg;
+	instance->tx_cfg_initialized = false;

kzalloc() will already set this to false.

regards,
dan carpenter