Re: forkables: About hardlink creation and NTFS transaction in rename()

[Michael Haubenwallner <michael dot haubenwallner at ssi-schaefer dot com>]

On 11/21/2016 04:19 PM, Michael Haubenwallner wrote:
> Hi Corinna,
> 
> now working with the cygfork patches (hardlinks to retain forkability
> beyond exe/dll update): when creating the forkable hardlink using the
> earlier opened file handle I may get STATUS_TRANSACTION_NOT_ACTIVE.
> 
> It turns out that when loaded 'some.dll' was readonly, then
> rename("new/some.dll", "some.dll") uses an NTFS-transaction to drop
> the readonly attribute, breaking the subsequent hardlink creation
> of "/var/run/cygfork/.../soname.dll" via the original file handle.

It turns out that when NTFS transactions are involved with the dll,
creating a hardlink requires the FILE_WRITE_ATTRIBUTES access.

Strange enough, this also applies when the file handle used to create
the hardlink is opened _after_ the rename transaction has finished.

On the other hand, opening the file handle with FILE_WRITE_ATTRIBUTES
access _before_ the rename(), the transactional unlink_nt fails with
STATUS_TRANSACTIONAL_CONFLICT.

> Now I can see these options:
> * As using NTFS transactions seems not to be recommended any more,
>   might we drop the NTFS transaction from rename() and _unlink_nt ()?
> * Or do I need to create an NTFS-transaction in dll_list::alloc()
>   when opening the original dll file handle?

some more now:

* Just for creating the hardlink open the file by id, determined
  in dll_list::alloc(), with immediately closing the file handle.
  However, I've not managed yet to open a file by id...

* When checking for, hardlinks are needed when the MemorySectionName
  resolves different than the (Get)ModuleFileName, and just for
  creating the hardlink open the file by current MemorySectionName.

> Thoughts?

Thanks!
/haubi/