staging: vc04_services: Need suggestions on trying to fix sparse warning in vchiq_arm.c

DISCUSSION Ojaswin Mujoo 3 participants 7 messages Jun 1, 2021 – Jun 6, 2021 Staging

Share & tools

Ojaswin Mujoo

author

2021-06-01 20:05 UTC #0 lore raw

Hello,

I was trying to address the following TODO item in "drivers/staging/vc04_services/interface/TODO" and would love some comments and suggestions on this:

Quoted 3 lines

14) Clean up Sparse warnings from __user annotations. See vchiq_irq_queue_bulk_tx_rx(). Ensure that the address of "&waiter->bulk_waiter" is never disclosed to userspace.

More specifically, I was looking at ways to fix the following sparse warning:

drivers/staging/vc04_services/interface/vchiq_arm/vchiq_arm.c:1001:26:

warning: incorrect type in assignment (different address spaces) expected void *[assigned] userdata
got void [noderef] __user *userdata

From my understanding, the issue seems to be that the (void *)userdata local variable in vchiq_irq_queue_bulk_tx_rx() can be assigned a (void __user *) or a (struct bulk_waiter *). This makes the assignment tricky since it can either be a userspace pointer or a kernel pointer.

Right now, we are just ignoring the sparse warning which is essentially resulting in the __user attribute being lost when we assign args->userdata to userdata. This can be dangerous as it might result in illegal access of userspace memory, without any sparse warning, down the line.

Further, this issue seems to boil down to the fact that the (void *)userdata field in struct vchiq_bulk (in vc04_services/interface/vchiq_arm/vchiq_core.h) can, again, either be a (stuct bulk_waiter *) or a (void __user *). To fix this, I was playing with the idea of modifying this userdata field of struct vchiq_bulk to be something like the following:

	diff --git a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_core.h 
	b/ drivers/staging/vc04_services/interface/vchiq_arm/vchiq_core.h

	--- a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_core.h
	+++ b/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_core.h

	@@ -227,10 +227,16 @@ enum vchiq_bulk_dir {
	 
	 typedef void (*vchiq_userdata_term)(void *userdata);
	 
	+struct vchiq_userdata_or_waiter {
	+		void __user *userdata;
	+		struct bulk_waiter *bulk_waiter;
	+               
	+};
	+
	 struct vchiq_bulk {
			short mode;
			short dir;
	-		void *userdata;
	+		struct vchiq_userdata_or_waiter *userdata_or_waiter;
			dma_addr_t data;
			int size;
			void *remote_data;

I was then planning on modifying all the code that works with vchiq_bulk->userdata accordingly. I believe this can help us overcome the sparse warnings, as well as preserve the __user attribute in cases when the userspace pointer does get assigned here.

However, since I'm not very familiar with the codebase, I just wanted to confirm if this is an acceptable workaround and to make sure I'm not breaking anything or overlooked anything here. I noticed that we also want to make sure that bulk_waiter's address is not exposed to userspace. Will it be possible to provide some pointers on how/where this might happen, so I can see if I can try to extend this patch to avoid that.

I would love to hear you suggestions and thoughts on this.

PS: I'm new here so please do correct me incase I missed anything.

Regards,
Ojaswin

Dan Carpenter

2021-06-01 20:23 UTC #1 lore raw

The problem is not the Sparse warning, the problem is that this code is a mess. It used to very clearly buggy and I reported the bug. I think Arnd found the bug again independently and fixed it.

A couple weeks ago Al Viro looked at this code. Here is his write up:

https://www.spinics.net/lists/kernel/msg3952745.html

It shouldn't take Al Viro dozens of pages of detailed analysis to try figure out if the code is safe or not. Your idea silences the warning but would make the code even more subtle and complicated.

The right thing to do is to re-write the code to be simpler.

regards,
dan carpenter

Ojaswin Mujoo

author

2021-06-02 14:50 UTC #2 lore raw

Quoted 8 lines · Dan Carpenter

The problem is not the Sparse warning, the problem is that this code is a mess. It used to very clearly buggy and I reported the bug. I think Arnd found the bug again independently and fixed it.
A couple weeks ago Al Viro looked at this code. Here is his write up:
https://www.spinics.net/lists/kernel/msg3952745.html

Quoted 9 lines

It shouldn't take Al Viro dozens of pages of detailed analysis to try figure out if the code is safe or not. Your idea silences the warning but would make the code even more subtle and complicated.
The right thing to do is to re-write the code to be simpler.
regards,
dan carpenter

Thank you for the prompt reply and the link, it was very insightful. You are right, I was definitely going about this the wrong way and missing the larger picture. I'll spend some time trying to understand this codebase as I think that'd be a good start to understand how stuff works in the kernel (even though some of the things in this driver are anti patterns) and hopefully get some ideas on ways to clean this up.

Anyways, thanks again for the help, cheers!

Ojaswin

Stefan Wahren

2021-06-04 06:13 UTC #3 lore raw

Hi Ojaswin,

Am 02.06.21 um 16:50 schrieb Ojaswin Mujoo:

Quoted 18 lines · Dan Carpenter

Quoted 8 lines
The problem is not the Sparse warning, the problem is that this code is a mess. It used to very clearly buggy and I reported the bug. I think Arnd found the bug again independently and fixed it.
A couple weeks ago Al Viro looked at this code. Here is his write up:
https://www.spinics.net/lists/kernel/msg3952745.html
Quoted 9 lines
It shouldn't take Al Viro dozens of pages of detailed analysis to try figure out if the code is safe or not. Your idea silences the warning but would make the code even more subtle and complicated.
The right thing to do is to re-write the code to be simpler.
regards,
dan carpenter

Quoted 8 lines

Thank you for the prompt reply and the link, it was very insightful. You are right, I was definitely going about this the wrong way and missing the larger picture. I'll spend some time trying to understand this codebase as I think that'd be a good start to understand how stuff works in the kernel (even though some of the things in this driver are anti patterns) and hopefully get some ideas on ways to clean this up.
Anyways, thanks again for the help, cheers!

thanks for your interest in cleaning this up. Yes, it's not clear which points on the TODO list are the lower hanging fruits. In case you don't want to fix checkpatch issues, maybe you can look at points 8, 9, 10, 12 and 13. Most of them require testing with a Raspberry Pi, but feel free to ask if you have problems with it.

Regards
Stefan

Quoted 7 lines

Ojaswin
_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Ojaswin Mujoo

author

2021-06-05 07:23 UTC #4 lore raw

Quoted 30 lines · Stefan Wahren

Hi Ojaswin,
Am 02.06.21 um 16:50 schrieb Ojaswin Mujoo:
Quoted 18 lines · Dan Carpenter
Quoted 8 lines
The problem is not the Sparse warning, the problem is that this code is a mess. It used to very clearly buggy and I reported the bug. I think Arnd found the bug again independently and fixed it.
A couple weeks ago Al Viro looked at this code. Here is his write up:
https://www.spinics.net/lists/kernel/msg3952745.html
Quoted 9 lines
It shouldn't take Al Viro dozens of pages of detailed analysis to try figure out if the code is safe or not. Your idea silences the warning but would make the code even more subtle and complicated.
The right thing to do is to re-write the code to be simpler.
regards,
dan carpenter
Quoted 8 lines
Thank you for the prompt reply and the link, it was very insightful. You are right, I was definitely going about this the wrong way and missing the larger picture. I'll spend some time trying to understand this codebase as I think that'd be a good start to understand how stuff works in the kernel (even though some of the things in this driver are anti patterns) and hopefully get some ideas on ways to clean this up.
Anyways, thanks again for the help, cheers!

Quoted 9 lines

thanks for your interest in cleaning this up. Yes, it's not clear which points on the TODO list are the lower hanging fruits. In case you don't want to fix checkpatch issues, maybe you can look at points 8, 9, 10, 12 and 13. Most of them require testing with a Raspberry Pi, but feel free to ask if you have problems with it.
Regards
Stefan

Got it, Task 10 (cdev to its own file) seems like a pretty good task to get started with. I'm planning to buy a Rpi 4 so I think I can run tests on that.

Thank you so much for the help, I'll get back incase I face any issues down the line.

Regards,
Ojaswin

Quoted 8 lines

Quoted 7 lines
Ojaswin
_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Stefan Wahren

2021-06-05 08:41 UTC #5 lore raw

Hi Ojaswin,

Am 05.06.21 um 09:23 schrieb Ojaswin Mujoo:

Quoted 36 lines · Stefan Wahren

Quoted 35 lines
Hi Ojaswin,
Am 02.06.21 um 16:50 schrieb Ojaswin Mujoo:
Quoted 18 lines · Dan Carpenter
Quoted 8 lines
The problem is not the Sparse warning, the problem is that this code is a mess. It used to very clearly buggy and I reported the bug. I think Arnd found the bug again independently and fixed it.
A couple weeks ago Al Viro looked at this code. Here is his write up:
https://www.spinics.net/lists/kernel/msg3952745.html
Quoted 9 lines
It shouldn't take Al Viro dozens of pages of detailed analysis to try figure out if the code is safe or not. Your idea silences the warning but would make the code even more subtle and complicated.
The right thing to do is to re-write the code to be simpler.
regards,
dan carpenter
Quoted 8 lines
Thank you for the prompt reply and the link, it was very insightful. You are right, I was definitely going about this the wrong way and missing the larger picture. I'll spend some time trying to understand this codebase as I think that'd be a good start to understand how stuff works in the kernel (even though some of the things in this driver are anti patterns) and hopefully get some ideas on ways to clean this up.
Anyways, thanks again for the help, cheers!
thanks for your interest in cleaning this up. Yes, it's not clear which points on the TODO list are the lower hanging fruits. In case you don't want to fix checkpatch issues, maybe you can look at points 8, 9, 10, 12 and 13. Most of them require testing with a Raspberry Pi, but feel free to ask if you have problems with it.

Quoted 6 lines

Quoted 3 lines
Regards
Stefan
Got it, Task 10 (cdev to its own file) seems like a pretty good task to get started with. I'm planning to buy a Rpi 4 so I think I can run tests on that.

okay, but the AFAIK the vchiq driver in the mainline kernel doesn't work with Rpi 4 yet. The Raspberry Pi 3 B Plus is currently the recommend devel platform, so you can test 32 and 64 bit kernel.

Best regards

Quoted 17 lines

Thank you so much for the help, I'll get back incase I face any issues down the line.
Regards,
Ojaswin
Quoted 6 lines
Quoted 6 lines
Ojaswin
_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Ojaswin Mujoo

author

2021-06-06 09:26 UTC #6 lore raw

Quoted 46 lines · Stefan Wahren

Hi Ojaswin,
Am 05.06.21 um 09:23 schrieb Ojaswin Mujoo:
Quoted 36 lines · Stefan Wahren
Quoted 35 lines
Hi Ojaswin,
Am 02.06.21 um 16:50 schrieb Ojaswin Mujoo:
Quoted 18 lines · Dan Carpenter
Quoted 8 lines
The problem is not the Sparse warning, the problem is that this code is a mess. It used to very clearly buggy and I reported the bug. I think Arnd found the bug again independently and fixed it.
A couple weeks ago Al Viro looked at this code. Here is his write up:
https://www.spinics.net/lists/kernel/msg3952745.html
Quoted 9 lines
It shouldn't take Al Viro dozens of pages of detailed analysis to try figure out if the code is safe or not. Your idea silences the warning but would make the code even more subtle and complicated.
The right thing to do is to re-write the code to be simpler.
regards,
dan carpenter
Quoted 8 lines
Thank you for the prompt reply and the link, it was very insightful. You are right, I was definitely going about this the wrong way and missing the larger picture. I'll spend some time trying to understand this codebase as I think that'd be a good start to understand how stuff works in the kernel (even though some of the things in this driver are anti patterns) and hopefully get some ideas on ways to clean this up.
Anyways, thanks again for the help, cheers!
thanks for your interest in cleaning this up. Yes, it's not clear which points on the TODO list are the lower hanging fruits. In case you don't want to fix checkpatch issues, maybe you can look at points 8, 9, 10, 12 and 13. Most of them require testing with a Raspberry Pi, but feel free to ask if you have problems with it.
Quoted 6 lines
Quoted 3 lines
Regards
Stefan
Got it, Task 10 (cdev to its own file) seems like a pretty good task to get started with. I'm planning to buy a Rpi 4 so I think I can run tests on that.

Quoted 5 lines

okay, but the AFAIK the vchiq driver in the mainline kernel doesn't work with Rpi 4 yet. The Raspberry Pi 3 B Plus is currently the recommend devel platform, so you can test 32 and 64 bit kernel.
Best regards

Hey Stefan,

Ahh got it. sure I can get a 3B+ instead, thanks for the heads up.

Quoted 10 lines

Quoted 10 lines
Thank you so much for the help, I'll get back incase I face any issues down the line.
Regards,
Ojaswin
Quoted 2 lines
Quoted 2 lines
Ojaswin

Quoted 9 lines

Quoted 8 lines
Quoted 4 lines
Quoted 4 lines
_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Regards,
Ojaswin