Skip to contentSkip to conversation

keys/keyctl: Use kfree_rcu instead of kfree

PATCH In review Siddh Raman Pant 4 participants 8 messages Jul 23, 2022 – Jul 28, 2022

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 keys/keyctl: Use kfree_rcu instead of kfree

In keyctl_watch_key, use kfree_rcu() for freeing watch and wlist as they support RCU and have an rcu_head in the struct definition.

---
 security/keys/keyctl.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
security/keys/keyctl.c +2 -2
index 96a92a645216..087fbc141cfd 100644
@@ -1832,9 +1832,9 @@ long keyctl_watch_key(key_serial_t id, int watch_queue_fd, int watch_id)
 	}
 
 err_watch:
-	kfree(watch);
+	kfree_rcu(watch, rcu);
 err_wlist:
-	kfree(wlist);
+	kfree_rcu(wlist, rcu);
 err_wqueue:
 	put_watch_queue(wqueue);
 err_key:
Signature · 7 lines
2.35.1


_______________________________________________
Linux-kernel-mentees mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees
#1 lore raw
Quoted 2 lines · Siddh Raman Pant via Linux-kernel-mentees

In keyctl_watch_key, use kfree_rcu() for freeing watch and wlist as they support RCU and have an rcu_head in the struct definition.

That does not explain why this change is needed. What problem does this solve? Why use RCU if you don't have to? What functionality did you just change in this commit and why?

And how was this tested?

thanks,

greg k-h

#2 lore raw
Quoted 3 lines · On Sat, 23 Jul 2022 19:35:16 +0530 Greg KH <[email protected]> wrote:

That does not explain why this change is needed. What problem does this solve? Why use RCU if you don't have to? What functionality did you just change in this commit and why?

We can avoid a race condition wherein some process tries to access them while they are being freed. For instance, the comment on watch_queue_clear() also states that:
/*
* Remove all the watches that are contributory to a queue. This has the
* potential to race with removal of the watches by the destruction of the
* objects being watched or with the distribution of notifications. */
And an RCU read critical section is initiated in that function, so we should use kfree_rcu() to not unintentionally free it while it is in the critical section.

Quoted 1 line

And how was this tested?

It compiles locally for me, and I used syzbot on this along with testing the other watch_queue_clear patch, which generated no errors.

Thanks,
Siddh
_______________________________________________
Linux-kernel-mentees mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees

#3 lore raw
Quoted 5 lines · Siddh Raman Pant
Quoted 3 lines · On Sat, 23 Jul 2022 19:35:16 +0530 Greg KH <[email protected]> wrote:

That does not explain why this change is needed. What problem does this solve? Why use RCU if you don't have to? What functionality did you just change in this commit and why?

Quoted 11 lines

We can avoid a race condition wherein some process tries to access them while they are being freed. For instance, the comment on watch_queue_clear() also states that:
/*
* Remove all the watches that are contributory to a queue. This has the
* potential to race with removal of the watches by the destruction of the
* objects being watched or with the distribution of notifications. */
And an RCU read critical section is initiated in that function, so we should use kfree_rcu() to not unintentionally free it while it is in the critical section.

You need to explain all of this in a changelog text. Don't say what you do, but say why you are doing it.

Quoted 4 lines
Quoted 1 line

And how was this tested?

It compiles locally for me, and I used syzbot on this along with testing the other watch_queue_clear patch, which generated no errors.

How does the watch queue stuff relate to this keyctl logic?

Again, be specific as to why you are doing things.

thanks,

greg k-h

#5 lore raw
In reply to Siddh Raman Pant
Quoted 16 lines · Siddh Raman Pant
Quoted 3 lines · On Sat, 23 Jul 2022 19:35:16 +0530 Greg KH < [email protected]> wrote:

That does not explain why this change is needed. What problem does this solve? Why use RCU if you don't have to? What functionality did you just change in this commit and why?

We can avoid a race condition wherein some process tries to access them while they are being freed. For instance, the comment on watch_queue_clear() also states that:
/*
* Remove all the watches that are contributory to a
queue. This has the
* potential to race with removal of the watches by the
destruction of the
* objects being watched or with the distribution of
notifications.

Quoted 4 lines
         */

And an RCU read critical section is initiated in that function, so we should use kfree_rcu() to not unintentionally free it while it is in the critical section.

That doesn't apply in this case, does it? watch and wlist are locally allocated and neither has been made externally visible if the error leg is taken, so they should just be locally freed, which is what the code was doing before this proposed patch.

James

_______________________________________________
Linux-kernel-mentees mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees

#4 lore raw
In reply to Greg KH
Quoted 2 lines · On Sat, 23 Jul 2022 20:13:11 +0530 Greg KH <[email protected]> wrote:

You need to explain all of this in a changelog text. Don't say what you do, but say why you are doing it.

Okay, I will keep this in mind next time.

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

And how was this tested?

It compiles locally for me, and I used syzbot on this along with testing the other watch_queue_clear patch, which generated no errors.

How does the watch queue stuff relate to this keyctl logic?

Again, be specific as to why you are doing things.

It doesn't relate, I just wanted to say that syzbot didn't crash too (I had this change in the same branch as that patch for testing, and syzbot compiled it successfully).

Sorry for the confusion.

Though now as James has pointed out, this patch isn't needed.

Apologies,
Siddh
_______________________________________________
Linux-kernel-mentees mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees

#6 lore raw
In reply to James Bottomley
Quoted 4 lines · On Sat, 23 Jul 2022 20:20:49 +0530 James Bottomley…

That doesn't apply in this case, does it? watch and wlist are locally allocated and neither has been made externally visible if the error leg is taken, so they should just be locally freed, which is what the code was doing before this proposed patch.

You are correct.

Sorry for this, I should have looked at it for a tad bit more.

Thanks,
Siddh
_______________________________________________
Linux-kernel-mentees mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees

maintainer
#7 lore raw
In reply to Siddh Raman Pant
Quoted 4 lines · Siddh Raman Pant

In keyctl_watch_key, use kfree_rcu() for freeing watch and wlist as they support RCU and have an rcu_head in the struct definition.

Applies to any patch: the commit message should *clearly* describe

1. What is wrong in the current code *behaviour*.
2. Why does the code change save the day.

Quoted 24 lines
---
 security/keys/keyctl.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
security/keys/keyctl.c +2 -2
index 96a92a645216..087fbc141cfd 100644
@@ -1832,9 +1832,9 @@ long keyctl_watch_key(key_serial_t id, int watch_queue_fd, int watch_id)
 	}
 
 err_watch:
-	kfree(watch);
+	kfree_rcu(watch, rcu);
 err_wlist:
-	kfree(wlist);
+	kfree_rcu(wlist, rcu);
 err_wqueue:
 	put_watch_queue(wqueue);
 err_key:
Signature · 1 line
2.35.1

BR, Jarkko