On Wed, Aug 31, 2016 at 6:22 AM, Greg Kroah-Hartman
<gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> On Thu, Aug 18, 2016 at 08:00:59PM -0700, Arve Hjønnevåg wrote:
>> On Thu, Aug 18, 2016 at 7:30 PM, ZhaoJunmin Zhao(Junmin)
>> <zhaojunmin@xxxxxxxxxx> wrote:
>> >
>> >
>> > 在 2016/8/18 23:23, Greg Kroah-Hartman 写道:
>> >>
>> >> On Tue, Aug 16, 2016 at 07:44:59PM -0700, Arve Hjønnevåg wrote:
>> >>>
>> >>> On Mon, Aug 15, 2016 at 7:58 AM, Greg Kroah-Hartman
>> >>> <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>> >>>>
>> >>>> On Thu, Jun 16, 2016 at 12:45:33AM +0200, Jann Horn wrote:
>> >>>>>
>> >>>>> If /dev/binder is opened and the opener process then e.g. calls execve,
>> >>>>> proc->vma_vm_mm will still point to the location of the now-freed
>> >>>>> mm_struct. If the process then calls ioctl(binder_fd, ...), the
>> >>>>> dangling
>> >>>>> proc->vma_vm_mm pointer will be compared to current->mm.
>> >>>>>
>> >>>>> Let the binder take a reference to the mm_struct to avoid this.
>> >>>>>
>> >>>>> v2: use the right refcounter
>> >>>>>
>> >>>>> Fixes: a906d6931f3ccaf7de805643190765ddd7378e27
>> >>>>
>> >>>>
>> >>>> Nit, the proper way to do this is:
>> >>>>
>> >>>> Fixes: a906d6931f3c ("android: binder: Sanity check at binder ioctl")
>> >>>>
>> >>>> You can get that by doing:
>> >>>> git show -s --abbrev-commit --abbrev=12 --pretty=format:"%h
>> >>>> (\"%s\")%n" a906d6931f3ccaf7de805643190765ddd7378e27
>> >>>>
>> >>>> Also, I'm guessing this should go to the stable kernels that include the
>> >>>> above patch?
>> >>>>
>> >>>
>> >>> Has this patch been tested?
>> >
>> >
>> > Hi:
>> > The patch have tested in huawei device.
>> >
>> >>>
>> >>> Does it actually fix anything?
>> >>>
>> >>> Changes to vma_vm_mm are normally protected by the mmap_sem it
>> >>> contains, and is set to NULL in the vma->vm_ops->close call on the
>> >>> binder vma. The vma_vm_mm pointer is only used to check if the mm
>> >>> struct that the binder driver was mmapped into is the same mm struct
>> >>> that the driver just locked (in binder_update_page_range). I don't
>> >>> think change description here is accurate. If a process calls execve,
>> >>> the old vma should be closed (and vma_vm_mm cleared) before the
>> >>> mm_struct that it belongs too is freed.
>> >>
>> >>
>> >> Yeah, this whole thing feels wrong...
>> >
>> >
>> > In huawei device,
>> > For example: system_server process use many libs,libstagefright,
>> > Libdrmframework etc.
>> > In these lib, binder global object exist like static sp<DeathNotifier>
>> > sDeathNotifier, static sp<IDrmManagerService> sDrmManagerService;
>> > When system_sever process fork failed, the new child process call exit,
>> > which will call hook destruction function.
>> > In destruction, the parent process *binder reference count will be wrong*.
>> > To protect
>> > these exception, we need to add the code in binder.
>>
>> I not sure I follow this.
>>
>> I see now that there is another patch that got merged that reused the
>> vma_vm_mm in the ioctl itself, without holding the lock that protects
>> it in the existing code. It would probably be best to revert that
>> change too.
>
> So what other patch should I revert? Have a git commit id?
>
> thanks,
>
a906d6931f3ccaf7de805643190765ddd7378e27
That patch appears to fix a user-space bug in the kernel, so I don't
think it is needed.
--
Arve Hjønnevåg
_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
