Re: Question: consult patch

From: zhangwensheng (E)
Date: Fri Aug 12 2022 - 03:39:08 EST


Hi Lee

I'm sorry I didn't fully understand you. did you mean the CVE
was caused by a commit which was applied to an internal kernel
version?  What commitment can be displayed? and I know that
mainline do not have the problem because of some code refactoring,
But are there any problems with some other stable branches?

thank you very much!
Wensheng

在 2022/8/12 15:26, Lee Jones 写道:
On Fri, 12 Aug 2022, zhangwensheng (E) wrote:

Hi

In CVE list last week, there is a new cve reported in asop 4.14 like below:
Reference link: https://lore.kernel.org/all/CAODzB9rgMexvLjE=WuTm+SN8SfUggaZgWG-aBcy6cotppju6mw@xxxxxxxxxxxxxx/T/

---
CVE-2022-20158: mm: backing-dev: Take a reference to the bdi in use to
prevent UAF

CVSS v3 score is not assigned.

AOSP kernel 4.14 contains following 2 patches.
- 69e8f03c5ced3e4e6fb4181f4dac185104e3420b ("mm: backing-dev: Take a
reference to the bdi in use to prevent UAF")
- 80d91b86a199798ee2321a0ab0f09e6e12764678 ("fs: explicitly unregister
per-superblock BDIs")

The first commit 69e8f03("mm: backing-dev: Take a reference to the bdi
in use to prevent UAF") is not merged in the mainline and stable
kernels.
Commit 80d91b8 was merged in 5.16-rc1(commit hash is
0b3ea0926afb8dde70cfab00316ae0a70b93a7cc) which requires commit
c6fd3ac ("mm: export bdi_unregister") that exports symbol of
bdi_unregister().

Fixed status
mainline: [0b3ea0926afb8dde70cfab00316ae0a70b93a7cc]
---

As mentioned above, patch 69e8f03c5ced ("mm: backing-dev: Take a
reference to the bdi in use to prevent UAF") in asop 4.14 can fix
a null dereference problem, form my analysis, may like below:

blk_cleanup_queue
    blk_put_queue
        kobject_put(&q->kobj)
                blk_release_queue
                    blk_exit_queue
                        bdi_put
                            release_bdi  // bdi -> null
del_gendisk
    bdi_unregister(disk->queue->backing_dev_info) // null -> reference

From my analysis, In asop 4.14 kernel, in loop_remove function, there is
such a timing that executing "blk_cleanup_queue" first and then
"del_gendisk".
but because of the refcnt of queue will add by hte line
"WARN_ON_ONCE(!blk_get_queue(disk->queue));" in "device_add_disk", which may
not result in "bdi_put" releasing bdi  in "blk_cleanup_queue".

I'm not sure where the problem is, so I want to ask Lee Jones who sent this
patch
for clarification.

Thanks!

Wensheng

在 2022/8/12 13:44, Christoph Hellwig 写道:
On Fri, Aug 12, 2022 at 11:34:59AM +0800, zhangwensheng (E) wrote:
Hi Lee :
    I saw your patch because of CVE-2022-20158, the patch like below:

---
    mm: backing-dev: Take a reference to the bdi in use to prevent UAF
I can't see that patch anywhere, and I've not seen an bug report for it.

    Because of a distinct lack of locking and/or reference taking,
    blk_cleanup_queue() puts the final taken reference to the bdi, which
.. and blk_cleanup_queue also is gone upstream.

What am I missing?
The issue reported in the aforementioned CVE was caused by a commit
which was applied to an internal, device specific repository. One
that has never existed in Mainline. I failed to reproduce the KASAN
report in any upstream or stable tree without the offending patch
applied. The issue was fixed in all affected internal trees.