Messages in this thread | | | Date | Sun, 3 Feb 2008 21:21:35 +0300 | From | Oleg Nesterov <> | Subject | Re: [PATCH] sys_remap_file_pages: fix ->vm_file accounting |
| |
(remove stable@kernel.org from CC)
On 02/02, Matt Helsley wrote: > > On Wed, 2008-01-30 at 20:26 +0300, Oleg Nesterov wrote: > > > > Offtopic. I noticed this problem while looking at this patch: > > > > http://marc.info/?l=linux-mm-commits&m=120141116911711 > > > > So this (the old vma could be removed before we create the new mapping) > > means that the patch above has another problem: if we are remapping the > > whole VM_EXECUTABLE vma, removed_exe_file_vma() can clear ->exe_file > > while it shouldn't (Matt Helsley cc'ed). > > > > Oleg. > > Looking at sys_remap_file_pages() it appears that the shared flag must > be set in order to remap. Executable mappings are always MAP_PRIVATE and > hence lack the shared flag so that any modifications to those areas > don't get written back to the executable. I don't think userspace can > change this flag
Yes, userspace can't change it. But if MVFS changes ->vm_file it could also change vm_flags... But I think you are right anyway, we shouldn't care.
So I have to try to find another bug ;) Suppose that ->load_binary() does a series of do_mmap(MAP_EXECUTABLE). It is possible that mmap_region() can merge 2 vmas. In that case we "leak" ->num_exe_file_vmas. Unless I missed something, mmap_region() should do removed_exe_file_vma() when vma_merge() succeds (near fput(file)).
Oleg.
| |