[-next,0/5] fs: fix possible null-ptr-deref when parsing param

Message

Hawkins Jiawei Oct. 23, 2022, 4:39 p.m. UTC

According to commit "vfs: parse: deal with zero length string value",
kernel will set the param->string to null pointer in vfs_parse_fs_string()
if fs string has zero length.

Yet the problem is that, when fs parses its mount parameters, it will
dereferences the param->string, without checking whether it is a
null pointer, which may trigger a null-ptr-deref bug.

So this patchset reviews all functions for fs to parse parameters,
by using `git grep -n "\.parse_param" fs/*`, and adds sanity check
on param->string if its function will dereference param->string
without check.

Hawkins Jiawei (5):
  smb3: fix possible null-ptr-deref when parsing param
  nfs: fix possible null-ptr-deref when parsing param
  ceph: fix possible null-ptr-deref when parsing param
  gfs2: fix possible null-ptr-deref when parsing param
  proc: fix possible null-ptr-deref when parsing param

 fs/ceph/super.c      |  3 +++
 fs/cifs/fs_context.c | 58 +++++++++++++++++++++++++++++++++++++++++++-
 fs/gfs2/ops_fstype.c | 10 ++++++++
 fs/nfs/fs_context.c  |  6 +++++
 fs/proc/root.c       |  3 +++
 5 files changed, 79 insertions(+), 1 deletion(-)

Comments

Al Viro Oct. 23, 2022, 4:48 p.m. UTC | #1

On Mon, Oct 24, 2022 at 12:39:41AM +0800, Hawkins Jiawei wrote:
> According to commit "vfs: parse: deal with zero length string value",
> kernel will set the param->string to null pointer in vfs_parse_fs_string()
> if fs string has zero length.
> 
> Yet the problem is that, when fs parses its mount parameters, it will
> dereferences the param->string, without checking whether it is a
> null pointer, which may trigger a null-ptr-deref bug.
> 
> So this patchset reviews all functions for fs to parse parameters,
> by using `git grep -n "\.parse_param" fs/*`, and adds sanity check
> on param->string if its function will dereference param->string
> without check.

How about reverting the commit in question instead?  Or dropping it
from patch series, depending upon the way akpm handles the pile
these days...

Hawkins Jiawei Oct. 24, 2022, 12:42 a.m. UTC | #2

On Mon, 24 Oct 2022 at 00:48, Al Viro <[e-mail]> wrote:
>
> On Mon, Oct 24, 2022 at 12:39:41AM +0800, Hawkins Jiawei wrote:
> > According to commit "vfs: parse: deal with zero length string value",
> > kernel will set the param->string to null pointer in vfs_parse_fs_string()
> > if fs string has zero length.
> >
> > Yet the problem is that, when fs parses its mount parameters, it will
> > dereferences the param->string, without checking whether it is a
> > null pointer, which may trigger a null-ptr-deref bug.
> >
> > So this patchset reviews all functions for fs to parse parameters,
> > by using `git grep -n "\.parse_param" fs/*`, and adds sanity check
> > on param->string if its function will dereference param->string
> > without check.
>
> How about reverting the commit in question instead?  Or dropping it
> from patch series, depending upon the way akpm handles the pile
> these days...

I think both are OK.

On one hand, commit "vfs: parse: deal with zero length string value"
seems just want to make output more informattive, which probably is not
the one which must be applied immediately to fix the
panic.

On the other hand, commit "vfs: parse: deal with zero length string value"
affects so many file systems, so there are probably some deeper
null-ptr-deref bugs I ignore, which may take time to review.

Ian Kent Oct. 24, 2022, 3:34 a.m. UTC | #3

On 24/10/22 08:42, Hawkins Jiawei wrote:
> On Mon, 24 Oct 2022 at 00:48, Al Viro <[e-mail]> wrote:
>> On Mon, Oct 24, 2022 at 12:39:41AM +0800, Hawkins Jiawei wrote:
>>> According to commit "vfs: parse: deal with zero length string value",
>>> kernel will set the param->string to null pointer in vfs_parse_fs_string()
>>> if fs string has zero length.
>>>
>>> Yet the problem is that, when fs parses its mount parameters, it will
>>> dereferences the param->string, without checking whether it is a
>>> null pointer, which may trigger a null-ptr-deref bug.
>>>
>>> So this patchset reviews all functions for fs to parse parameters,
>>> by using `git grep -n "\.parse_param" fs/*`, and adds sanity check
>>> on param->string if its function will dereference param->string
>>> without check.
>> How about reverting the commit in question instead?  Or dropping it
>> from patch series, depending upon the way akpm handles the pile
>> these days...
> I think both are OK.
>
> On one hand, commit "vfs: parse: deal with zero length string value"
> seems just want to make output more informattive, which probably is not
> the one which must be applied immediately to fix the
> panic.
>
> On the other hand, commit "vfs: parse: deal with zero length string value"
> affects so many file systems, so there are probably some deeper
> null-ptr-deref bugs I ignore, which may take time to review.

Yeah, it would be good to make the file system handling consistent

but I think there's been a bit too much breakage and it appears not

everyone thinks the approach is the right way to do it.


I'm thinking of abandoning this and restricting it to the "source"

parameter only to solve the user space mount table parser problem but

still doing it in the mount context code to keep it general (at least

for this case).


Ian

Tetsuo Handa Oct. 31, 2022, 11:28 a.m. UTC | #4

On 2022/10/24 12:34, Ian Kent wrote:
> 
> On 24/10/22 08:42, Hawkins Jiawei wrote:
>> On Mon, 24 Oct 2022 at 00:48, Al Viro <[e-mail]> wrote:
>>> On Mon, Oct 24, 2022 at 12:39:41AM +0800, Hawkins Jiawei wrote:
>>>> According to commit "vfs: parse: deal with zero length string value",
>>>> kernel will set the param->string to null pointer in vfs_parse_fs_string()
>>>> if fs string has zero length.
>>>>
>>>> Yet the problem is that, when fs parses its mount parameters, it will
>>>> dereferences the param->string, without checking whether it is a
>>>> null pointer, which may trigger a null-ptr-deref bug.
>>>>
>>>> So this patchset reviews all functions for fs to parse parameters,
>>>> by using `git grep -n "\.parse_param" fs/*`, and adds sanity check
>>>> on param->string if its function will dereference param->string
>>>> without check.
>>> How about reverting the commit in question instead?  Or dropping it
>>> from patch series, depending upon the way akpm handles the pile
>>> these days...
>> I think both are OK.
>>
>> On one hand, commit "vfs: parse: deal with zero length string value"
>> seems just want to make output more informattive, which probably is not
>> the one which must be applied immediately to fix the
>> panic.
>>
>> On the other hand, commit "vfs: parse: deal with zero length string value"
>> affects so many file systems, so there are probably some deeper
>> null-ptr-deref bugs I ignore, which may take time to review.
> 
> Yeah, it would be good to make the file system handling consistent
> but I think there's been a bit too much breakage and it appears not
> everyone thinks the approach is the right way to do it.
> 
> I'm thinking of abandoning this and restricting it to the "source"
> parameter only to solve the user space mount table parser problem but
> still doing it in the mount context code to keep it general (at least
> for this case).

No progress on this problem, and syzbot is reporting one after the other...

I think that reverting is the better choice.

Ian Kent Nov. 1, 2022, 12:32 a.m. UTC | #5

On 31/10/22 19:28, Tetsuo Handa wrote:
> On 2022/10/24 12:34, Ian Kent wrote:
>> On 24/10/22 08:42, Hawkins Jiawei wrote:
>>> On Mon, 24 Oct 2022 at 00:48, Al Viro <[e-mail]> wrote:
>>>> On Mon, Oct 24, 2022 at 12:39:41AM +0800, Hawkins Jiawei wrote:
>>>>> According to commit "vfs: parse: deal with zero length string value",
>>>>> kernel will set the param->string to null pointer in vfs_parse_fs_string()
>>>>> if fs string has zero length.
>>>>>
>>>>> Yet the problem is that, when fs parses its mount parameters, it will
>>>>> dereferences the param->string, without checking whether it is a
>>>>> null pointer, which may trigger a null-ptr-deref bug.
>>>>>
>>>>> So this patchset reviews all functions for fs to parse parameters,
>>>>> by using `git grep -n "\.parse_param" fs/*`, and adds sanity check
>>>>> on param->string if its function will dereference param->string
>>>>> without check.
>>>> How about reverting the commit in question instead?  Or dropping it
>>>> from patch series, depending upon the way akpm handles the pile
>>>> these days...
>>> I think both are OK.
>>>
>>> On one hand, commit "vfs: parse: deal with zero length string value"
>>> seems just want to make output more informattive, which probably is not
>>> the one which must be applied immediately to fix the
>>> panic.
>>>
>>> On the other hand, commit "vfs: parse: deal with zero length string value"
>>> affects so many file systems, so there are probably some deeper
>>> null-ptr-deref bugs I ignore, which may take time to review.
>> Yeah, it would be good to make the file system handling consistent
>> but I think there's been a bit too much breakage and it appears not
>> everyone thinks the approach is the right way to do it.
>>
>> I'm thinking of abandoning this and restricting it to the "source"
>> parameter only to solve the user space mount table parser problem but
>> still doing it in the mount context code to keep it general (at least
>> for this case).
> No progress on this problem, and syzbot is reporting one after the other...
>
> I think that reverting is the better choice.

Yes, I agree/


Ian